Democracy, Accountability, and Representation Edited by Adam Przeworski ‘New York University | Susan C. Stokes University of Chicago Bernard Manin New York University | a CAMBRIDGE UNIVERSITY PRESSBernard Manin, Adam Przeworski, and Susan C. Stokes Chapter One Elections and Representation The claim connocting democracy and representation is that under democracy governments are representative because they are elected: if elections are freely contested, if participation is widespread, and if citizens enjoy political liberties, then gov- ernments will act in the best interest of the people. In one - the “mandate” - view, elections serve to select good policies or policy- bearing politicians. Parties or candidates make policy proposals during campaigns and explain how these policies would affect citizens’ welfare; citizens decide which of these proposals they want implemented and which politicians to charge with their implementation, and governments do implement them. Thus, elections emulate a direct assembly and the winning platform becomes the “mandate” that the government pursues. In a second = “accountability” ~ view, elections serve to hold governments responsible for the results of their past actions. Because they anticipate the judgment of voters, governments are induced to choose policies that in their judgment will be positively evaluated by citizens at the time of the next election. Yet both views are problematic. Representation is an issue because politicians have goals, interests, and values of their own, and they know things and undertake actions that citizens cannot observe or can monitor only at a cost. Even if once they are in office politicians may want to do nothing but serve the public, to get elected in the first place they may have to gratify special interests. And once elected, they may want to pursue their private goals or some public objectives that differ from those of citizens. If they have such motivations, they will want to do things other than represent the public. And voters do not know everything they need to know, whether to decide prospectively what politiciansManin, Praeworski, and Stokes should be doing or to judge retrospectively if they did what they hould have done. And if voters know that there are some things they do not know, they do not want to bind politicians to implement their wishes. In turn, if citizens do not have sufficient information to evaluate the incumbent governments, the threat of not being reelected is insufficient to induce governments to act in the best interest of the public. In this chapter we analyze whether voters can enforce repre- sentation by using their vote to choose policies and politicians, using it to sanction the incumbent, or using their vote simul- taneously in both ways. We then discuss institutional features that may be conducive to inducing representation. ‘The Mandate Conception of Representation and offer ve that politicians are not all the same, hey may attempt to secure representation by using their vote © choose best policies or policy-bearing politicians. The questions we need to examine are whether (1) electoral campaigns are formative, that is, voters can justifiably expect that parties would 4o what they proposed, and (2) pursuing the winning platform, the ‘mandate,” is always in the best interest of voters. We will say that ‘mandate-representation” occurs if the answer to both these {uestions is positive, that is, parties truthfully inform voters about heir intentions and the implementation of these intentions is best or voters under the given circumstances. ‘The mandate conception of representation is widespread: cholars, journalists, and ordinary citizens rely on it as if it were jomatic. Keeler (1993), for example, explains the major policy eforms introduced by Reagan, Thatcher, and Mittorand as allows: their respective countries faced economic crises, voters vanted change and expressed this desire at the polls, and the espective governments implemented their mandates. This model eems to account well for policy formation in advanced industrial ociety (Klingeman, Hofferbert, and Budge 1994). As a French olitician put it, “Since the Romans, it is an old law of politics, ‘hich we should never lose from our view: governments can last, uly by the principle by which they were born” (Séguin 1997). A rudimentary conceptual apparatus may help clarify what is atailed. In elections, parties or candidates present themselves to 30 Elections and Representation voters, informing them about their policy intentions." Specifically, they tell voters which policies they intend to pursue, for what purposes, and with which consequences. Once elected, the forious candidates choose policies, not necessarily the same as announced. Having observed the outcomes of the policies, voters vote again. To introduce an example, suppose that there are two possible platforms: S (for economic “security”) and E (for “efliciency”) policies.* Competing parties or candidates promise to do S or E, and once elected pursue S or E. Politicians may care both about policies and about being elected and reelected. Politicians have preferences over policies if their reward from holding office or the probability of r depends on the policies they adopt. One can think of the reward from holding office in three ways: politicians may have favorite policies and derive utility from implementing them, they may want to advance their private interests, or they may derive satisfaction from the honor attached to office. Politicians have some beliefs about the promises that are more likely to make them win and the policies that voters will in fact appreciate having experienced their results. ‘Thus, the question about mandate representation is whether (1) the policy of the incumbents will be the same as their electoral platform, and (2) whether pursuing this platform will be bes: for voters. The conditions under which mandate representation occurs are threefold: when politicians’ and voters’ interests oincide, when politicians are motivated by the desire to be reelected and they think that voters will reelect them if they pursue policies on which they campaigned, and when politicians are concerned about the credibility of their future promises. We discuss these situations in turn. 1. Interests of politicians coincide with those of voters.* s and governments have identical interests if govern- ‘ments want in their self-interest to bring about states of the world that are most desired by citizens. If politicians and voters "candidates alo extol hls personal virtues ope we tea later. = The terminology is ster (0999) » obvi, te question tat arises wit such formuladons fs, Which vor? We fecume in is chaplr cat, whatever the dimenslonaly of th swe spec, a Injriy rl ogullriam exists and this thre ens a“dedsve oer: Par oa pieatons that aise when iis asumpdon dacs not hla, se Feejon (1986, 1995) 31Manin, Przeworski, and Stokes also have the same beliefs about the effects of policies on outcomes (“technical beliefs” in the language of Austen-Smith 1990), then candidates get elected on the platform most preferred by voters and as incumbents they implement this platform in their own interest. Almost all discussions of representation, beginning with J. S. Mill’s Considerations on Representative Government (1991 (1861), assume that electing politicians who somehow mirror or reproduce the composition of the electorate achieves representation. The assembly is representative in this view if it is a miniature of the electorate, a sample of it. The hypothesis underlying this conviction is that if the assembly is descriptively representative, then it will act to represent interests of the represented. As a consequence, discussions of representative institutions focus almost exclusively on electoral systems (for an example, see Rogowski 1981). The pathbreaking, and still unduly ignored, contribution of Pitkin (1967) was to problematize this connection: is it true that propor- tionality is the best way to secure representation? If each repre- sentative puts forth opinions and promotes the interests of his constituency, will the best interest of the collectivity be served? And what if representatives become different from their coi stituents by the mere fact of being representatives? What if, once elected, they acquire knowledge the constituents do not have and perhaps even interests of their own? 2. Politicians want to be elected and reelected.* And they expect that voters will reelect them if they pursue the policies “In its pure frm. spatial theory of voting is logically incomplete: voters care only ‘ahout outcomes but they choose on the basis of policies. What is obviously missing, are “technical beliefs.” as defined earlier. Note that if candidates and voters ba Identical interests but different technical beliefs, they will have different induced preferences about polices, * One difficulty with this view, manifest in Ml, is that while the assembly may reflect interests proportionately, many decisions entailed in governing do not permit propertional allocations. Indeed, many are dichotomous, and in those the ajorty prevalls while the minority loses. Henee, while proportionality allows all Voices to be heard it does not guarantee that all interests will be proportionately accommodated. Thus, as Pitkin points out, the activity of representing entailed by descrip representation consists t most of articulating views, not of making "This is true regardless of whether politicians also have other interests as long as they puta high value on holding office per se 32 that they offered in their election campaign. If election-seeking politicians know the preference of the decisive voter, they offer a platform that coincides with this preference. If they expect that voters’ preferences will not change or that these prefer- ences will be confirmed by observing the outcomes of imple: menting the mandate (Harrington 1993a), then the incumbents pursue the announced policy in quest of reelection. And if voters know what is good for them, the outcome is best for voters, so that voters’ threat to punish deviations from the mandate is credible. 3. Politicians are concerned that their promises be credible in the future. Even if voters believe that a deviation from the mandate was beneficial for them, they discount future promises of politicians who acquire a reputation of reneging on their campaign promises. Hence, voters may threaten the incumbents to vote against them if they betray their promises, regardless of the outcomes. This threat is implemented by the rival party in Alesina’s (1988) model, in which voters are not strategic. In turn, it is implemented by voters in Banks's (1990) model, where, in turn, the credibility of this threat is assumed, rather than derived.’ Hence, this is at best an incomplete story. We return to later. Note that mandate representation is a situation in which policies adopted by incumbents follow their electoral platforms and these policies are the best for citizens under the conditions observed by the incumbents. The three possibilities distinguished earlier add up to the conclusion that mandate represemation occurs when what politicians and voters want coincides or when politicians care only about winning elections, and to win they must promise and implement policies that are best for the public. But short of this happy coincidence, politicians may have incentives either to deviate from the mandate in the best interest of the public or to stick to it at the cost of the electorate. * Banks justifies this assumption referring to the multiperiod model of Austen- Smith and Banks (1989), where the threat of punishing deviations Is indeed crediblo. But in Austen-Smith and Banks, governments never quite fulfill promises: when voters expect little of governments, parties always promise ‘more than they deliver, even though they deliver first best; in turn, when voters ‘expect a lot, platforms and reelection chances are independent of what vouers want. 33Manin, Przoworski, and Stokes ‘To highlight the weakness of the mandate mechanism, that politicians cannot be reelected even once.* Voters know that once elected the incumbent will do whatever he wants. Without the sanction of voting again, voters must guess which of the competing parties or candidates has policy preferences that with theirs and which is impervious to the corrupting sway of office. Yet unless the pool of candidates includes such politicians and unless voters guess correctly who they are, the victorious candidates will not act in the representative manner. If they have policy preferences distinet from those of the decisive voter, they will deviate from the announced policies; if they pursue private benefits, they will extract rents. Moreover, just to get elected, politicians may have to make promises to special interests. Suppose, in the spirit of the Chicago School of Regulation (Stigler 1975; Peltzman 1976; Becker 1958, 1983), that (1) voters are ignorant, rationally or not, about the impact of policies on their welfare, and (2) to present themselves to voters, politicians need to expend resources, including but not limited to money. Politicians are concerned only with winning elections, but to win they must raise resources. Because voters do not care about policies that have only a small impact on their welfare, politicians can sell to interest groups policies that inflict only a small cost on each individual voter but which concentrate benefits on the particular interest groups and spend on elec- tioneering the resources contributed by interest groups in exchange for these policies. Since policies that raise resources from special interests are costly to voters, politicians choose policies that make them indifferent at the margin between increasing voters’ welfare and campaign expenditures, and the welfare of voters is not maximized. To take an example, suppose that politicians decide whether to "In fact, itis enough that the number of elections in which « politician ean run is ‘known and finite. Suppose thatthe politician will aot be able to run after the t th term. Then during the (~ 1)st election voters will know that in the last term the politician wil have no incentives to soek reelection and will vole against him, But ifthe poltician will not be reelected for the t-th term, then he will not have incentives to behave well during the ((~ 1)st term and voters will not elect hin. But then the same will be true during the (2nd term, all the way to the first one. Unless politicians care about voters’ approval when they leave office for the last tino, term limits deprive voters from creating incentives for politicians to represent them. 34 Elections and Representation subsidize the sugar industry. By subsidizing the sugar industry, the government inflicts on each individual an annual cost of $5.75 and benefits the sugar industry to the tune of $1.5 billion. Voters will not want to spend their resources to learn about the sugar policy and its effects: this information costs more than $5.75. Then the government will subsidize, get a campaign contribution from the industry, and maximize its probability of reelection.” Indeed, according to the Center for Responsive Policy (New York Ties, January 24, 1997, p. 3), a sugar price subsidy that adds an extra $50 to a five-pound bag of sugar was supported by sixty-one senators who received on the average $13,473 from the industry political action committee, while it was opposed by thirty-five senators who on the average got $1,461. ‘The fact is that just to exist and to present themselves to voters, political parties must raise funds. When these funds come from special interests, they are exchanged for favors. Presumably, if Philip Morris Co. Inc. contributed in 1996 over $2.5 million to the Republican National Committee (New York Times, January 28, 1997, p. 3), it must have expected at least $2.5 million in favors; otherwise, its management should have been thrown out by stockholders. Exchanges of political contributions for policy favors torting through their effects on the allocation of resources, And the social cost of such distortions is likely to be much greater than that of outright thefi, which is distorting only through its effect on distorting taxes Yet situations in which either politicians deviate from their promises in voters’ best interest or stick to them against the interests of a majority are possible even if incumbents face repeated elections and even if electoral campaigns are costless (or publicly funded). 1. Conditions may change in such a way that the imple- mentation of the mandate is no longer best for voters. Suppose ° This argument is subject to two criticisms. One is that voters are only raticnally ignorant, & la Stigler (1975), governments will be constrained to limit such policies to those that inflict a small cast on voters: hence, the aggregate less of welfare may not be very large. Cleaely what matters is how gullible voters are, land Becker (1983), who drops the assumption that ignorance is only rational, offers not even an intuition about it. Second (soe Arnold 1993 and our subsequent discussion), there are several groups and most importantly the partisan opposition that have the interest wo diffuse for free information about such policies, 35Manin, Przeworsi Stokes that immediately upon assuming office an incoming government that won the election campaigning for policy S learns sometl neither it nor the voters knew at the time of the electioi the departing government, competing for reelection, hid the sad contents of the treasury, and the electoral victors assuming office that coffers are empty. Say that S policy when conditions are good, while E is better when they are bad. Then the government faces the choice of deviating from the mandate in the best interest of voters or adhering to it in spite of the changed circumstances. In turn, voters, who would have agreed to changing the policy had they observed what the government observes, must decide whether to believe the government's message, which will appear self-serving, without direct access to this information. And they can err, in either direction. Hence, incumbents will deviate some of the time and voters punish some deviations, good or bad. Note that even changes of conditions that are endogenous to government policy, but were unforeseen by politicians before they reached office, may be reasons to change course in the interest of citizens’ welfare; the 1983 switch of the French Socialist government may be a case in point. . To be elected, a candidate must offer the platform preferred by the decisive voter. Suppose a candidate believes that the decisive voter has incorrect beliefs about the effect of policies on outcomes. This candidate then faces a choice of offering a platform that she thinks is better for voters and going down to defeat (perhaps hoping to win the next time around, if the competitor implements less effective policies) or proposing what voters want and having at least a 50-50 chance of winning (if the other competitor offers the same platform). If the incumbent believes that the less popular policy is sufficiently more effective than the ‘one voters prefer, he or she anticipates that, having observed its effects, voters will become persuaded that the correet policy was chosen and will vote to reelect, so that the politician will be able to continue the policy that is in effect better for citizens. ‘Two situations need to be further distinguished. In one (Harrington 1993a: sec. 4), the two candidates have the same beliefs about the preferences of the decisive voter but different beliefs about the effectiveness of policies, meaning that one of them thinks the decisive voter is wrong. Then they offer the same platform in the electoral campaign but once elected may pursue 36 Elections and Representation different policies. If the winner is the candidate who believes that the policy preferred by the decisive voter is significantly inferior to the alternative, the incumbent adopts the policy she prefers, in the belief that voters will be persuaded about its superiority once the outcomes materialize. In such situations, we should observe candidates offering the same platform and then some:imes deviating from the mandate. In the second situation, the two candidates have the same beliefs about the effectiveness of policies but differ in their beliefs about the preference of the decisive voter. If elected, they pursue the same policy, but to get elected they offer different platforms. If the winner is the candidate who believed that the voters are mistaken, the incumbent switches policies once elected. In such situations, we should observe candidates offering diffe platforms and then pursuing the same policy regardless of who is elected. Note that in both of these situations candidates (may) deviate from their platforms once elected, but they deviate believing that they are acting in the best interest of the electorate. 3. Suppose that everything is the same as just described but the incumbent does not believe that voters will be persuaded by the effectiveness of the better policy ~ either because voters are quite certain which policy is better for them or because the policy choice does not make a great difference. As Harrington (1993a) shows, if voters initially believe that one policy is better than the other, they are harder to convince ex post of the superiority of the alternative. Fearing that if he proposes one platform and pursues another he will not be reelected, the incumbent offers the inferior platform that voters prefer and implements it, against what sare citizens’ best interests. The mandate will be implemented, ians will not act in a representative ‘manner. ‘To summarize: under some conditions, incumbents may either pursue policies that enhance that welfare of voters by deviating “The intuition is the following. Suppose that voters intaly believe that policy S fs better for them than E by some amount ¢. If they observe an outcame of implementing E that is botter than S by the amount e, their posterior bebe will, be that E produces outcomes at a level between their Intal expectations and their observation, which wil stil be below S. To be persuaded that E is better than S, the outcome of E would have to be better than the prior on S by more than e. 37Manin, Przeworski, and Stokes from the mandate or they may adhere to the mandate even if they think that implementing it is not best for voters. And if implementing the mandate is not the best the government can do, then the threat of punishing incumbents who deviate from it is not credible. Voters may not like governments that betray promises, but they will not punish politicians who made them sufficiently better off by deviating from the mandate. This impunity is mitigated by reputational considerations (Downs 1957; Ferejohn 1995). Politicians may be concerned about adhering to promises as an investment in credibility. Indeed, the Polish government was said to “be forced to remind itself of the promises of 1993 and to make some concessions to the voters, under the penalty of losing its credibility” (Krauze 1994). If incumbents anticipate that voters will not only look at their past policies but also pay attention to their new promises, that is, if their past performance in office is not fully informative, they must be concerned about being believed, which, in turn, moderates their temptation to deviate from the old promises. A politician who executed a pirouette will have to rely solely on his past performance when seeking reelection, while a politician who stuck to promises will be more likely to be believed next time around. In turn, voters may want to punish politicians who renege on their promises as an investment in information. After all, voters want their choices to have consequences; hence, they want to be able to predict the behavior of politicians from their campaign platforms, rhetoric, or identity. Politicians may claim that un- foreseen circumstances are the reason they deviated from their mandates. But they have some explaining to do, which is prima facie evidence that they think they are expected to follow mandates, While such reputational mechanisms may encourage the incumben's to adhere to electoral promises, a striking feature of democratic institutions, highlighted by Manin (1997), is that politicians are not legally compelled to abide by their platform in any democratic system. In no existing democracy are repre- sentatives subject to binding instructions. Citizens’ suits against governments that betrayed specific campaign promises have been cted by courts in several countries, most recently in Poland. No national-level democratic constitution allows for recall, and, except for the U.S. House of Representatives, electoral terms tend to be long - on the average, 3.7 years for legislatures and 3.9 years 38 Elections and Repres for presidents (Cheibub and Przeworski, Chapter 7 in this volume). While provisions for impeachment and procedures for with- drawing confidence are common, they are never targeted at the betrayal of promises." Binding national referenda based on initiative are found only in Switzerland and, in more ive forms, in Italy and Argentina. Hence, once citizens elect representatives, they have no institutional devices to force them to adhere to promises. Voters can sanction deviations from mandates only after their effects have been experienced Why then are there no institutional mechanisms to force officeholders to be faithful to their platforms? Historically, the main argument was that legislatures should be allowed to deliberate. People want their representatives to learn, one from another. Moreover, when people are uncertain about their judgments, they may want representatives to consult experts. ‘Another historical argument was that voters may not trust their own judgments. People not only may be afraid of their own passions but, if they are rationally ignorant, they must know that they do not know. Presumably, elections establish the calendar for when the accounts are to be taken. Hence, citizens may want to give the government some latitude to govern and evaluate government's actions at election times. O'Flaherty (1990) argues that this is a reason to elect politicians for fixed terms; in this way citizens can guard themselves against inconsistent time preferences and yet exercise ex post control. Finally, institutions must allow for changing conditions. No electoral platform can specify ex ante what the government should do in every contingent state of nature; governments must have some flexibility in coping with changing circumstances. If citizens expect that conditions may change and governments are likely to be representative, they will not want to bind governments by their instruetions."® Hence, there are good reasons why democratic institutions Occasionally @ deviation from mandates provides part of the impetus for Impeachment, even though deviation is not the formal justification. Two recent presidents who abandoned thelr campaign promises. in Venezuela and Ecuador, were impeached, one immediately, with no time allowed for the outcomes 10 materialize. "= Minford (1995: 105) observes In the context of monetary policy that “if voters have litle information, they may prefer w let governments have complete discretion, regardless of the lack of credibility, rather than te thelr hands.” 39Mani 1. Przeworski, and Stokes contain no mechanisms enforcing adherence to mandates. We choose policies that represent our interests or candidates who Tepresent us as persons, but we want governments to be able to govern. AS a result, while we would prefer governments to stick to their promises, democracy contains no institutional mech- anisms that insure that our choices would be respected. ‘The Accountability Conception of Representation Even if citizens are unable to control governments by obliging them to follow mandates, citizens may be able to do so if they can induce the incumbents to anticipate that they will have to render accounts for their past actions. Governments are “accountable” if voters can discern whether governments are acting in their interest and sanction them appropriately, so that those incumbents who act in the best interest of citizens win reelection and those who do not lose them. Accountability representation occurs when (1) voters vote to retain the incumbent only when the incumbent acts in their best interest, and (2) the incumbent chooses policies necessary to get reelected. To understand why the problem of accountability arises, we must consider again politicians’ objectives. Politicians may want to do nothing that well-informed citizens would not have wanted them to do; they may be public-spirited and dedicate themselves fully to furthering the public interest. But they may also want something different from and costly to citizens, whether just some goals that citizens do not share, reelection, or private gains. Politicians may want to pursue their own ideas even if these differ from those of citizens.” Some may care most about advancing their careers against fellow politicians, within the government or the same party. Some may seek perks (Niskanen 1971). Some may want to get rich at the expense of citizens, while in office or after leaving it. Some may be most concerned about recognition by foreigners. In all these cases politicians will want something whose pursuit is injurious to citizens. For a lack of a better term and to keep with the standard terminology, we will refer to this something as “rents.” ‘To introduce another term standard in literature, there are ™ Suppose that in a poor country people want to consume immediately while benevolent politicians want to develop the country by increasing investment 40 Elections and Representation different ways in which politicians can “shirk,” that is, do things that citizens would not want them to do. They shirk if they spend time conspiring against their rivals. They shirk if they act to increase their own wealth. They shirk if they extend clientilistic favors to their families and friends. But the most important way in which they can act against the best interests of their con- stituents is by choosing policies that advance their own interests or the interests of some special interests to which they are beholden. ‘The problem facing citizens is then to set up a trade-off for politicians — between extracting rents and losing office or not extracting rents and staying in office ~ that would induce them to keep rents low, where keeping rents low may mean just doing what voters want. The standard view of how tho accountability mechanism operates relies on “retrospective voting.” In this view, citizens set some standard of performance to evaluate gov- ernments, such as “My income must increase by at least 4 percent during the term,” “Streets must be safe,” or even “The national team must qualify for the World Cup.” They vote against the incumbent unless these criteria are fulfilled. In turn, the government, wanting to be reelected and anticipating the citizens’ decision rule, does whatever possible to satisfy these criteria. Imagine that the conditions under which the government makes decisions can be “good” or “bad.” Governments decide whether to implement policy S, which is better for citizens when conditions are good, or policy E, which is better when conditions are bad. Suppose that the rents the incumbents obtain when they do all they can for the public consist just of their salaries and legally authorized perks of office, and suppose that the incumbents value being reelected. To make this analysis less abstract, examine a numerical example in which the legally qualified rents equal r* 1 +e, where ¢ is some small number, and the value of being reelected is V = 2. Let the structure of payoffs be as follows (the first numbar in each pair represents government rents, but citizens observe only their welfare, which is the second number): Government Implement $ Implement £ “Good” 1+e5 33 Conditions “Bad” a1 1+e3 4and Stokes Suppose now that the electorate knows everything it needs to know: Then, to induce politicians to act as well as they can under the circumstances, voters set their reolection rule as “When conditions are good, vote for the incumbent if the outcome is at least 5. When conditions are bad, vote for the incumbent if the outcome is at least 3; otherwise throw the rascals out.” A government facing good conditions knows that by choosing $ it will get r* = 1+ e and it will be certainly reelected, thus obtaining r* +V> 3, waere 3 is the most it can get by choosing E and not being reelected. In turn, a government facing bad conditions knows that by choosing E it will get r* + V > 3, which is what it would get by adopting S. Hence, the government acts in a representative manner and citizens get the most they can under either conditions. Accountability induces representation (Key 1966). As Fiorina (1981: 11) put it: “Given political actors who fervently desire to retain their positions and who carefully anticipate public reaction to their records as a means to that end, a retrospective voting electorate will enforce electoral accountability, albeit in an ex post, not an ex ante, sense.” Yet suppose that voters do not know what the conditions are. Politicians know these conditions. but voters may be unable to observe them at all or they may be able to monitor them only at a cost. Such conditions may include the negotiating posture of foreign governments or international financial institutions (something citizens cannot observe) or the level of demand in the major recipients of the country’s exports (something voters can observe only if they turn into economists). Then voters are in a quandary. if they set the standard the incumbent must meet at 5 and conditions turn out to be bad, the incumbent cannot be reelected whatever he does and he will seek excess rents. In turn, ifvoters set the standard at 3, the incumbent will be able to extract excess rents when the conditions happen to be good and be reelected by giving voters less than he could have given them, Whatever voters decide to do, politicians will sometimes escape from their control. One aspect of incomplete information merits particular attention. Note that the voters who animated the previous pages ‘were myopic: they were concerned only with the change of their welfare during the current term. But if voters are fully rational, they should also care at the end of the term about the present value of their future welfare: the legacy the incumbent leaves 42 Elections and Representation for the future. If the economy grows because the government cuts all the trees in the country, the voter will live on champagne during the term, but there will be no trees left to cut. In turn, if the economy declined because it underwent structural reforms, voters will have suffered economic deprivation but may have improved their life chances for the future. Yet all that voters observe is the change of welfare during the term, and they have to make inferences about the future on this basis. Say that voters observe that their current welfare declines: should they infer that the government is investing in their future or pursuing some (neoliberal) chimeras of its own or just robbing them blind? Following Stokes (1996a), note that voters can adopt one of three postures: 1, They can extrapolate the present experience into the future, This is the “normal” posture, insofar as this is what models of retro- spective voting normally assume. 2, They can assume an “intertemporal” posture (Przeworski 1996), expecting that the worse things have gotten, the better they augur for the future. 3. They can assume an “exonerating” posture, attributing thi of their welfare to bad conditions, rather than to anything the government did, decline It is hard to tell what is rational for people to do under these circumstances. Some empirical work on neoliberal reforms in new democracies (Przeworski 1996; Stokes 1996b) suggests that people are willing to exonerate governments for inflation and to treat increases of wages intertemporaily, as forecasting inflation, but that they are risk-averse about unemployment and turn against governments that generate it, Yet other studies come to the conclusion that people are sensitive to inflation and relatively indifferent to unemployment (Rose 1997; Weyland 1996). In any case, there is little on which people can base these judgments. Accountability models of elections assume typically that while voters do not know something that they need to know to evaluate governments, incumbents do know what they need to do in order to be reelected. The implicit artifice on which these models rely is that voters offer a contract to the government: “if you give us at least this, we will vote for you: otherwise, we will not.” Yet voters do not offer such contracts. Note that we could cast not only ballots but also a list of our conditions for reolecting the incumbent. But 43Manin, Przoworski, and Stokes we do not, and we do not because we want governments to do all they can for us, rather than just fulfill our minimal demands Indeed, Manin (1997) points out that voters can decide whether to Teelect the incumbent on any basis they want, including qualifying for the World Cup, and that they can change their mind between the beginning and the end of a term. At least in this way, voters are sovereign. Hence, a question arises about how the incumbent will act if information is asymmetric both ways: when voters are not cer- tain about the conditions under which policies are made but incumbents are uncertain what would satisfy voters. It can be shown (Cheibub and Przeworski, Chapter 7 in this volume) that when incumbents very much care about being reelected, they will represent, always pursuing policies appropriate to the conditions they observe, so that voters are better off keeping their expectations secret. When, however, incumbents care less about being reelected and, in turn, voters expect that conditions are likely to be good, votors are better off formulating exacting demands and making them known to politicians, Finally, when incumbents attach less value to being reelected and voters expect conditions to be bad, there is nothing voters can do to prevent incumbents from extracting excess rents when conditions happen {0 be good. Hence, voters are better off if they can reveal or not reveal their demands strategically. But to do so, they must still know how much the incumbent cares about being reelected and how likely it is that conditions are good. In sum, accountability is not sufficient to induce representation when voters have incomplete information Using the Vote for Two Purposes In a pure accountability model, voters use the vote only for one purpose, which is to sanction the incumbent, and the entire information available to voters is revealed by the performance ‘cumbent. In a pure mandate model, voters compare Promises candidates make about the future, and use the vote only to choose the better candidate. In Downs's (1957; also Fiorina 1981) model, voters use the information about the past per- formance of the incumbent and, if available, of challengers, but this is also a mandate model in the sense that voters use the information about the past only to choose a better government for 44 Elections and Representation the future, Indeed, Sniderman, Glaser, and Griffin (1990) claim that purely retrospective voting would be irrational: rational people look forward. Yet this is not right: if voters can credibly employ their vote only to sanction the incumbent, threatening to Use it this way is a perfectly rational way of inducing governments to act well in the future. For all we know, voters do not meditate whether to use the one instrument they have, the vote, to choose a better government or to structure incentives for incumbents. Fearon (Chapter 2 in this volume) offers persuasive stories to the effect that voters want to select good policies and politicians. Yet voting “to keep them honest” seems also ingrained in the repertoire of the democratic culture. The fact remains that voters have only one instrument to reach two goals: to select better policies and politicians, and to induce them to behave well while in office. The question then is what happens when voters try to use the vote for both purposes. ‘Suppose that, believing that politicians are not all the same, voters are swayed by the prospect of electing better governments. Voters may believe that the challenger is more competent, having a better understanding of the relation between policies and outcomes, or is more honest, being willing to accept lower rents in exchange for holding public office. An election takes place, an incumbent is installed, both the incumbent and the electorate observe the objective conditions, voters set their voting rule, incumbents choose rents, and voters vote again. Having observed what happened during the term, voters vote for a challenger with the probability that the challenger will be better than the incumbent (in the sense that she would have generated higher welfare under the same conditions). Anticipating that voters will vote for the challenger with some positive probability, the incumbent will then require a higher level of rents. Thus the rents that are necessary to induce the incumbent to seek reelection when voters use the vote as a selection device are larger than in the case when voters are only concerned about incentives for the incumbent. Using the vote to choose a better government prospectively is costly to voters in terms of their control over the incumbent (see Fearon, Chapter 2 in this volume). Note immediately that the following nightmare may arise (Perejohn 1986; Banks and Sundaram 1993): if voters always think that the challenger is better, then the incumbent can never be reelected, and he will always choose to extract high rents. In turn, 45Manin, Przeworski, and Stokes ections and Representation incumbents extract high rents, voters will never vote for them. The incumbent knows that votors will always be swayed by the promises of the challenger and always extract maximal rents, Which means that if voters believe that politicians are not all the same, they are certain that the challengers will be better for them, In this situation, voters’ control breaks down completely."* Yet, while voters may be gullible, they cannot be that ingenuous. The performance of the incumbent is informative. As Bartels (1988) discovered, in the United States at least, the past performance of a president is a good predictor of his future per- formance (and not of a challenger’s). Thus, voters who use their Vote prospectively have good reasons to rely on retrospective information. Harrington (1993b) shows that the more uncertain voters are about the effect of policies on outcomes, the more they should rely on information about past performance. They can observe the past performance of the incumbent and then decide how likely itis that a challenger is better. Nevertheless, as long as voters use their vote to elect a better government, they must lower the power of incentives for the incumbent. Madison (Federalist no, 57) thought that “The aim of every political constitution is, or ought to be, first to obtain for rulers men who possess most wisdom to discern, and most virtue to pursue, the common good of the society; and in the next place, to take the most effectual precautions for keeping them virtuous whilst they continue to hold their public trust.” Using the vote for both purposes — to obtain the best rulers and to keep them virtuous = is not irrational: while voters lose some control over the incumbent, in exchange they elect a better government. Yet the system Madison and his colleagues designed makes it possible to strive for one goal only at the expense of the other, Institutions, Elections, and Representation Democracies are not all the same, and itis possible that some democratic systems foster representation better than others. "This may som freee But several countries, notably Ecuador and Plan, experienced a series felons in which th challenger used e pure ap extansonar plc, was beeved by voters, swehed pon secu es trsctonary oe, to be defend by a challenger pramicng un eapansone policy, ete. " 46 While we have little systematic knowledge about the effects of particular institutional arrangements on voters’ control over politicians, some institutional factors merit attention. 1. Voters must be able to assign clearly the responsibility for government performance. Their ability to do so is limited when the government is a coalition. It is also limited when the pres- dency and the congress are controlled by different parties. It takes an elaborate theory of government to figure out who is responsible for what under such conditions (but see Anderson 1995). Hamilton argued in Federalist no. 70 that accountability is obscured under a plural, that is, cabinet executive: “But one of the weightlest objections to a plurality in the executive... is that it tends to conceal faults and destroy responsibility. ...The cir- cumstances which may have led to any national miscarriage or misfortune are sometimes so complicated that there are a number of actors who have different degrees and kinds of agency, though we may clearly see upon the whole that there has been mismanagement, yet it may be impracticable to pronounce to whose account the evil which may have been incurred is truly chargeable.” But a similar ambiguity arises in presidential systems, Bagehot (1992: 67) expressed this view most forcefully: “Two clever men never exactly agree about a budget. ++. They are sure to quarrel, and the result is sure to satisfy neither. And when the taxes do not yield as they were expected to yield, who is responsible? Very likely the secretary of the treasury ‘could not persuade the chairman - very likely the chairman could not persuade his committee - very likely the committee could not persuade the assembly. Whom, then, can you punish ~ whom can you abolish - when your taxes run short?” ‘The empirical findings concerning clarity of responsibility, most of them due to Powell and his collaborators, are confusing. Majority-inducing institutions increase the distance between the ideal position of the median voter and of the government (Huber and Powell 1996), but they increase what Powell (1990) measures as the “clarity of responsibility,” which, in turn, makes voting for incumbents more sensitive to economic performance (Powell and Whitten 1993). Hence, it seems that majoritarian institu- tions generate governments that are farther from voters in policy space but more accountable. The relation between citizens’ ences and the actual policies ~ not studied by Powell ~ is 47Manin, Preeworski, and Stokes 2. Voters must be able to vote out of office parties responsible for bad performance, and the parties they select must be able to enter government. These may appear to be universal features of democracy, but under some electoral systems they are next to impossible: witness the continued tenure of the Christian Democrats in Italy or of the LDP in Japan, or the weak connection between voting results and electoral outcomes in Bolivia. As Pasquino (1994: 25) put it with regard to Italy, “governing parties seemed to expropriate the voters of the political influence by making and unmaking governments at all levels with very little respect for electoral results.” 3. Politicians must have incentives to want to be reelected. This condition becomes problematic when there are limitations on reeligibility, ubiquitous in presidential systems (Cheibub and Przeworski, Chapter 7 in this volume), and when political parties fare not continuing bureaucratic organizations that offer their militants career prospects (Zielinski 1997). Paldam (1991) observed that the coefficients of the funetion relating the probability of reelection to economic outcomes are higher and their estimates are tighter when the party system is stable. 4. The opposition must monitor the performance of the government and inform citizens. Indeed. any reasonable under- standing of representation must include the opposition. Citizens have two agents, not just one: the incumbents who govern and the opposition that wants to become the government. The opposition is an agent of citizens because it wants to win office and, in order to win office, it must also anticipate the retrospective judgments that voters will make about the incumbents at election time. Anticipating these judgments, the opposition has incentives to monitor the government and to inform (truthfully or not) voters about the performance of the incumbents. Yet the existence of an opposition that wants to and can monitor government performance should not be taken for granted. Opposition can collude with the government'* or it can be so " Grain (1977) argued that in a single-member, single-district electoral system incumberts are unlikely to ever run against each other and that they therefore share an interest in ralsing barriers to entry of challengers from either party, Dasgupta (1993) offered another model of collusion and an argument that we should subsidize now parties. 48 Elections and Representation divided that it spends most efforts on internal fights rather than on the incumbents. The opposition may see no chance of winning and do something else rather than monitor the government: (see Pasquino 1994 with regard to the PCI in Italy). And it may or may not have resources to do so; among Powell’s (1990) twenty countries, only nine provide resources for the opposition in leg- islative committees. Yet, conversely, an opposition that always opposes is not any more credible to voters than the government. If every time the government says something, the opposition claims it is false, voters are not any better informed. Thus, the opposition plays a role in informing voters only when it neither always colludes with nor always contradicts the government. 5. The media, the role of which is emphasized by Arnold (1993), thus have a particular role to play. Unless they have clearly partisan interests, they are more credible than either the government or the opposition. 6. Finally, but perhaps most importantly, voters must have some institutional instruments to reward and punish governments for outcomes they generate in different realms. Yet elections are inherently a blunt instrument of control: voters have only one decision to make with regard to the entire package of government policies. Suppose that a government during one term has to make ten decisions, the incumbent makes all decisions against the interest of the majority, and the challenger offers to make one right decision. Then citizens will elect a government that will make nine wrong decisions. Obviously, the question is why would some other challenger not offer to make two right decisions, or three, or four, up to all ten. One answer is the barriers to entry: partisan politics is the most protected industry in the United States, Bur if there were no barriers to entry, then parties would have an incentive to form, promise to make all ten right decisions, make none right, and get out. If entrance into the electoral system entails fixed costs, competition will be limited; if itis free, then parties do not suffer when they lose. So we get either highly collusive party systems, such as the United States, or completely ephemeral systems, such as in Ecuador, where there is a new party system at each election.’ In either way, control is limited. "On the importance of electoral systems for the rent extraction by partes, see Meyerson (1993).Manin, Przeworski, and Stokes Conclusion: Elections as a Mechanism of Representation Although democracy may not assure representation, it is still plausible that democracy is more conducive to representation than, alternative regimes. Yet the conclusion of this analysis must be that citizens’ control over politicians is at best highly imperfect in most Cemocracies. Elections are not a sufficient mechanism to insure that governments will do everything they can to maximize citizens’ welfare. This is not an argument against democracy but one for insti- tutional reform and for institutional innovation. We need electoral institutions that enhance clarity of responsibility and make it easy for citizens to reward and punish those responsible. We need moral and economic conditions in which public service would enjoy respect as well as appropriate material rewards. In addition, we need institutions that would provide citizens with independent information about the government: “accountability agencies,” in the terms of an Australian Commission on Government Reform (Dunn and Ubr 1993). Such institutions may include (1) an independent board to assure transparency of campaign con- tributions," with its own investigative powers; (2) an independent auditing branch of the state, an auditor-general (World Bank 1994: 32), in the vein of the Chilean contraloria; (3) an independent source of statistical information about the state of the economy; and (4) a privileged place for the opposition in overseeing the publicly owned media. Yet even if responsibilities are clearly assigned, bad gov- ernments can be punished and good ones chosen, voters are well informed about the relations between politicians and special interests, and the rent-seeking behavior of politicians is well scrutinized, elections are just not a sufficient instrument of control over politicians. Governments make thousands of decisions that affect individual welfare; citizens have only one instrument to control these decisions: the vote. One cannot control a thousand targets with one instrument. Thus, for example, separating "A quesion of why private contributions should not be banned altogether naturally appears. Lalfont and Tirole (1994) argue that such a ban would be skirted by some politicians. If thi is rue and if money buys votes, then there Would fe an adverse selection process in which dishonest politicians would be jore likely to be elected. 50 Elections and Representation ‘monetary from other political decisions and voting separately, and at staggered intervals, for the directors of the central bank would give voters an additional instrument of control, with results superior both to giving discretion over monetary policy to the government and to delegating these decisions to a central bank independent from voters’ control (Minford 1995), The fact is that during the past two hundred years we have thought little about the institutional design of democracy. Since the great explosion of institutional thinking, when the present democratic institutions were invented ~ and they were invented ~ there has been almost no institutional creativity. Except for the never implemented provisions for workers’ comanagement in the Weimar Constitution, the discovery of proportional representation in the 1860s was the last major institutional invention, All democracies that have sprung up since the end of the eighteenth century, including the most recent ones, just combine in different ways, often piecemeal, the preexisting institutions. Hence, there is lots of room for institutional creativity. References 1, Alberto. 1988. “Credibility and Convergence in a Two-Party ‘System with Rational Voters.” American Economic Review 78 irom Dy ff Public Su t fe lorson, Christopher J. 1995. “The Dynamics of Public Support for vm palion Government” ComparatioePotealStades 28; 990-88 Arnold, Douglas. 1993, "Can Inattentive Citizens Control Their Elected Ropresentatives?” in Lawrence C. Dodd and Bruce I. Oppenheimer, cds., Congress Reconsidered, 401-16. 5th ed. Washington, D.C: Con- essional Quarterly Press. Austen Sh, Dud, 1990, “Creble Debate qulibria” Sot! Goce and Welfare 7: 75-93 ° Austen-Smith, David, and Jeffrey Banks. 1989, “Electoral Accountability and Incumbency.” In Peter C. Ordeshook, ed., Models of Strategic Ghoice in Politics, 121-50, Ann Arbor: University of Michigan Press. Bagehot, Walter. 1992. “The English Constitution: The Cabinet." In Arend Lijphart, ed. Parliamentary versus Presidential Government, 6-T1 Oxford: Oxford University Press. Banks, Jeffrey S. 1990, “A Model of Electoral Competition with Incomplete Information Joural of eon Theory 30: 209-28 janks, Joffrey S.. and Rangarajan K, Sundaram. 1993, “Adverse etbd Moral lizard n'a Repeated Tlectons Model In Wiliam A, Ales lection Benin, Przeworski, and Stoke: Barnett, Melvin J. Hinich, and Norman J. Schofield, eds.. Political Economy: Institutions, Competition, and Representation, 295-312. Cambridge: Cambridge University Press. Barro, Robert J. 1973. “The Control of Politicians: An Economic Model.” Public Choice 14: 19-42. Bartels, Larry. 1988. “The Economic Consequences of Retrospective Voting.” Department of Political Science, University of Rochester. Unpublished manuscript, Becker, Gary S. 1958. “Competition and Democracy.” Journal of Law and Economics 1: 105-9. 1983. “A Theory of Competition among Pressure Groups for Political Influence.” Quarterly Journal of Economics 98: 371-400. Crain, Mark W. 1977. "On the Structure and Stability of Political Markets.” Journal of Political Economy 85: 829-42. Dasgupta, Partha. 1993. An Inquiry into Well-Being and Destitution. Oxford: Clarendon Press. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row. Dunn, Delmer D., and John Uhr, 1993. “Accountability and Responsibility in Modern Democratic Governments.” Paper presented at the annual ‘meeting of the American Political Science Association, Washington, D.C., September 2-5. Elster, Jon. 1994. “The Impact of Constitutions on Economic Per- formance.” Proceedings of the World Bank Annual Conference on Development Economics, 209-26. Washington, D.C. Forejohn, John. 1986. “Incumbent Performance and Electoral Control.” Public Choice 50: 5-25, 1995. “The Spatial Model and Elections.” In Bernard Grofman, ed., Information, Participation, and Choice, 107-24. Ann Arbor: University of Michigan Press, Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. New Haven: Yale University Press, Hamilton, Alexander. 1982 [1788]. Federalist 70. In Alexander Hamilton, James Madison, and John Jay. The Federalist Papers, edited by Gary Wills, New York: Bantam, Harrington, Joseph E., Jr. 1993a. “The Impact of Reelection Pressures on the Fulillment of Campaign Promises.” Games and Economic Behavior 5: 71-97. 1993b. “Economic Policy, Economic Performance, and Elections.” American Economic Review 83: 27-42. Huber, John D., and G. Bingham Powell, Jr. 1996. “Congruence between Citizens and Policymakers in Two Visions of Liberal Democracy.” World Politics 49: 291-326. Keeler, John T. S. 1993. “Opening the Window for Reform: Mandates, 52 lect yns and Representation Crises, and Extraordinary Decision-Making.” Comparative Political Studies 25: 433-86. Key, V. 0, Jr. 1966, The Responsible Electorate. New York: Vintage. Klingoman, Hans-Dieter, Richard 1. Holferbert, and lan Budge. 1994. Parties, Policies, and Democracy. Boulder, Colo.: Westview Pros. Krauze, Jan, 1994, “La Pologne est menacée par immobilisme gou- vernemental.” Le Monde, September 19. Laffont, Jean-Jacques, and Jean Tirole. 1994. A Theory of Incentives in Procurement and Regulation. Cambridge, Mass.: MIT Press. Madison, James. 1982 [1788]. Federalist 57. In Alexander Hamilton, James Madison, and John Jay. The Federalist Papers, edited by Gary Wills. New York: Bantam. Manin, Bernard. 1997. Principles of Representative Government. ‘Cambridge: Cambridge University Press. Meyerson, Roger B. 1993. “Effectiveness of Electoral Systems for ‘Reducing Government Corruption.” Games and Economic Behavior 5: 118-32, ; Mill, John Stuart, 1991 [1861]. Considerations on Representative Government. Buffalo, N.¥.: Prometheus Press. Minford, Patrick. 1995. “Time-Inconsistency, Democracy, and Op:imal Contingent Rules.” Oxford Economic Papers 47: 195-210. Niskanen, William A. 1971. Bureaucracy and Representative Government. Chicago: University of Chicago Press. O'Flaherty, Brendan. 1990. “Why Are There Democracies? A Principal ‘Agent Answer.” Economics and Politics 2: 133-55. Paldam, Martin. 1991. “How Robust Is the Vote Function? A Study of Seventeen Nations over Four Decades.” In Helmuth Northop, Michael S. Lowis-Beck, and Jean-Dominique Lafay, eds., Economics and Politics: The Calculus of Support, 9-31. Ann Arbor: University of Michigan Press. Pasquino, Gianfranco. 1994. “Shaping a Better Republic? The Italian Case in a Comparative Perspective.” Working Paper No. 62. Madrid: Instituto Juan March de Estudios o Investigaciones. Peltzman, Sam. 1976. “Toward a More General Theory of Regulation.” Journal of Law and Economics 19: 209-87. Hanna F 1967. The Concept of Representation. Berkeley: University of California Press. Powell, G. Bingham, Jr. 1990. “Holding Governments Accountable: How Constitutional Arrangements and Party Systems Affect Clarity of Responsibility for Policy in Contemporary Democracies.” Paper presented at the annual meeting of the American Political Science Association, San Francisco. Powell, G. Bingham, Jr, and Guy Whitten. 1993. “A Cross-National rr 33Manir, Przeworski, and Stokes Analysis of Economic Voting: Taking Account of the Political Context.” American Journal of Political Science 37: 391-414. Preeworsii, Adam. 1996. “Public Support for Economie Reforms in Poland.” Comparative Political Studies 2%: 520-43, Rogowski, lonald, 1981. “Representation in Political Theory and in Law. Ethics 91: 395-430, Hose, Richard. 1997, “What Is the Demand for Price Stability in Post- Communist Countries?” University of Strathelyde Studies in Public Policy No. 282. Glasgow. ‘Séguin, Philippe. 1997. Liberation, May 29. Sniderman. Paul M., James M. Glaser, and Robert Griffin, 1990, “Information and Electoral Choice.” In John A. Ferejohn and James H. Kuklinski, eds., Information and Democratic Processes, 117-35. Urbana: University of linois Press. Stigler, George J. 1975, The Chicago: University of Chicago Press. Stokes, Susan C. 1996a. “Public Opinion and Market Reforms: The Limits of Economic Voting.” Comparative Political Studies 29; 499-519, 1996b, “Economic Reform and Public Opinion in Peru, 1990-1995." Comparative Political Studies 29: 544-65, Weyland, Kurt. 1996. “Risk Taking in Latin American Economic Restructuring: Lessons from Prospect Theory.” International Studies Quarterly 40: 185-208, World Bank. 1994, Governance: The World Bank's Experience, Washington, D.C.: World Bank. Zielinski, Jakub. 1997, “Democratic Consolidation: A Kole of Political Parties as Institutions of Accountability.” Paper presented at the annual meeting of the American Political Science Association, Washington, D.C. 54 James D. Fearon Chapter Two Electoral Accountability and the Control of Politicians: Selecting Good Types versus Sanctioning Poor Performance ‘The concept of accountability is not by itself problematic, or at least it should not be. We say that one person, A, is accountable to another, B, if two conditions are met. First, there is an under- standing that A is obliged to act in some way on behalf of B. Second, B is empowered by some formal institutional or perhaps informal rules to sanction or reward A for her activities or pe formance in this capacity. I this sense, employees are accountable to their employers, CEOs to their boards and their boards to stockholders, department chairs to the departments they represent, and elected politicians to their electorates. In the jargon of economic theory, relations involving accountability are agency relationships in which one party is understood to be an “agent” who makes some choices on behalf of a “principal” who has powers to sanction or reward the agent.’ Most interesting questions about accountability in political and economic contexts concern not its definition but rather the understanding of what activities or performance the agent is For valuable comments I wish wo thank the editors ofthe Volume and John Ferejohn ‘and Jim Snyder. ‘one might define & weaker notion of “accountability” in which one person is accountable to another if itis understood thatthe frst has a responsibilty to act 6 behalf ofthe second, independent of whether the second has sanctioning or ‘rewarding instruments. I think this blurs accountability with moral responsibility ‘and does not square with ordinary usage. which typically Involves the pre- sumption of monitoring and sanctioning instruments. 55James D. Fearon accountable for, the nature of the principal’s sanctioning or rewarding instruments, and the problem of to what extent a given system of incentives will lead the agent to act on behalf of the principal, that is, to do what the principal would want. In addition, in the case of electoral accountability, additional problems arise from the presence of multiple principals (voters, but perhaps also courts and other elected officials in some cases) rather than a single principal or a collective body that can act as a single principal, For instance, with multiple principals the question of saying what the principals would want can be difficult, even the- oretically, as Arrow’s theorem suggests. ‘There is an important prior question, however, about whether elections are best thought of in terms of accountability at all Certainly, an important tradition in democratic theory under- stands elections as mechanisms of political accountability. In view, elections are seen as a sanctioning device that induces elected officials to do what the voters want. The anticipation of not being reelected in the future leads elected officials not to shirk their obligations to the voters in the present (Barro 1973; Ferejohn 1986; Fiorina 1981: Key 1966; Manin 1997). 1 begin by developing a four-step argument that questions whether elections are best conceived as mechanisms of accountability. The four steps may be summarized as follows First, voters need not see elections as mechanisms that establish accountability; instead, they might understand elections as oppor- tunities to choose a “good type” of political leader, one who would act on their behalf independent of reelection incentives. Second, if voters can distinguish good types to some degree, then it follows that electoral accountability is not necessary for elections 10 produce public policy that the principals (the voting public) want. Third, empirical observations about public opinion and elections in the United States (and probably other democracies) suggest that in fact voters think about elections more as opportunities to select good types than as mechanisms establishing accountability and, in consequence, good behavior by politicians. Fourth, this sort of understanding is not idiosyncratic or foolish, but may actually make a good deal of sense if one also believes that repeated elections as sanctioning devices have little ability to induce politicians to do what the voters would want done. Even if voters think about elections more in terms of selection 56 trol of Polit than sanctioning, these two understandings are by no means incompatible or mutually exclusive. Indeed, successfully selecting for good types implies sanctioning bad types, which gives bad types an incentive to appear as if they were good types. Thus, if the electorate tries to select good types, then bad types may moderate their policy choices in the direction of what the electorate wants, as in the classical theory of electoral accountability. But then, insofar as bad types become harder to distinguish from good types. selection is rendered problematic. This brief argument suggests that selection and sanctioning necessarily interact and that the interaction has a strategic component (at least in the behavior of politicians). Following the discussion of the four-step argument, I pre: simple game model that depicts three views of how elec generate policy outcomes desired by voters: pure sanctioning moral hazard,” or accountability), pure selection, and a mixed case suggested by the preceding paragraph. Consistent with the fourth point summarized earlier, | show that as the voters’ ability to monitor politicians becomes very poor, the mixed case approaches a problem of pure selection, although the voters’ ability to sort out good types on the basis of an incumbent's record also diminishes. Analysis of the mixed case also provides insight into the interaction between selection and sanctioning. For example, I show that although the electorate would like to commit to a retrospective voting rule to motivate self-interested politicians optimally, when it comes time to vote it makes sense for the electorate to focus completely on the question of type: which candidate is more likely to be principled and share the public’s preferences? The simple logic behind this result may help explain the empirical observation that voters tend to think about elections in terms of choosing a good type. Elections with No Expectation of Accountability There is no logical reason why elections must be understood as a part of a relationship of accountability or “agency.” For example, a group of people might understand elections as a means of selecting or conferring honor on the best or most distinguished person, This sort of understanding can take two forms, one connected with conferring or recognizing honor without any 87James D, Fearon instrumental purpose, and one concerned with selecting the best person with a notion that this person will of her own accord do what is best for the voters. ‘Thus, in the first case, the voters have no expectation what- soever that the elected official has a responsibility to act on behalf of the electorate (thus violating the first condition necessary for a rela- tionship to entail accountability). The election might be understood simply as a declaration of who in the group most deserves the honor of political authority. As a logical claim this is simply true, Empirically, Mark Kishlansky’s (1986) account of parliamentary elections in early modern England suggests that something like this actually occurred, and Max Weber (1978; 1112-30) had earlier claimed that premodern elections were about the acclamation and recognition of charisma, rather than the selection of a delegate or agent, Echoes of this view can also be found in the “Michigan model” of elections, which sees votes as affirmations of warm feelings for a candidate (Campbell etal. 1960). In the second case, the voters choose with an eye toward whether the elected official would act on their behalf, but still have no expectation that the anticipation of future elections was a device for giving the elected official an incentive to do so (thus iolating the second condition). Imagine, for instance, a system in which elected officials can serve only one term in office, so elections cannot serve as a sanctioning device to induce good per- formance by those elected. But if the voters think they are able to distinguish among types of candidates and that some types are more inclined to act of their own accord in the public interest, then they could still understand the elections as the fundamental mechanism of democratic governance. Lack of Eleetoral Accountability Need Not Imply Lack of Responsiveness This second case suffices to show that lack of electoral accountability need not imply public policy opposed to voter Preferences. Imagine again the system in which elected officials can serve only one term in office, and thus are not accountable since they cannot be sanctioned at the polls.® Such a system could * Of course, they may still be accountable before the law for criminal and civil violations, as is everyone else. 58 Control of Politicians still, in principle, produce public policy in accord with what the voters would want, if the voters are able to distinguish candidates who are simply the type who want to do in office what they would want done. Define a good type for a particular voter as a politician who (1) shares the voter's issue preferences, (2) has integrity, in that he or she is hard to bribe or otherwise induce to work against the voter's interests, and (3) is competent or skilled in discerning and implementing optimal policies for the voter. If voters are able to distinguish politicians along these three dimensions, then this lack of accountability need no e public what it wants. The conclusion: Electoral accountability is not in principle necessary for elections to produce responsive public policy.! How might voters distinguish between good and bad types? Voters have available a variety of signals and measures, which might be partitioned as follows: 1. General measures of the voter's welfare (“Am I better eff than I was four years ago?") or of the health of the economy or society (he “misery index.” the crime rate, ete.) that might allow inferences about the types of incumbent politicians. 2, More specific measures of “pork” delivered, policies chosen, or votes cast by incumbent representatives in prior terms in office: party afiation and location within the party. 3. Information about the personal character of a politician, derivable from all manner of sources, such as life history as reported by the media or the candidate, impressions of how the candidate speaks, body language, sense of humor, involvement in scandals, and ‘group affiliations of the candidate. 4. Policies advocated by the candidate or Incumbent in campaign or other political speeches. Note that this list includes both “prospective” measures such as campaign platforms and “retrospective” measures such as how the economy performed under an incumbent. If people are trying to select good types more than sanctioning to induce good behavior * 1 mean “responsive” either in the weak sense of Stokes (Le., preferred by some tmalority) oF something stronger (soe Stokes, Chapter 3 in this volume). In the ble choice erature on what determines how representatives vote on Bs in Congress, John Lott (1987) has argued similarly that voters are able to“soet out” olicials with similar policy preferences and thus generate outcomes they lke. 59D. Fearon in the future, then both retrospective and prospective information may be relevant to their decisions, even if it is noisy and often interpreted badly. (By contrast. only retrospective measures are informative in a pure sanctioning view of elections, since promises about future performance are not credible. Voters trying to select ood types from campaign speoches face the problem of noisy and misleading signals, but informative signaling is not a priori impossible.) How Do Voters Think about the Point of Elections? ‘That voters might understand elections in terms of selection rather than sanctioning is more than an abstract, theoretical point. This conception ~ elections as a means of selecting the type of leader who will act competently and faithfully in the public interest, independent of reelection incentives - is arguably much closer to the popular understanding in the United States (and probably other democracies as well) than is the understanding that thinks of future elections as a sanctioning device to induce appropriate behavior during an elected official’s present term. We can see this by considering how the “elections as selecting a good type” perspective makes sense of a number of striking facts about public opinion and elections in the United States that are perplexing if one adopts an “elections as mechanisms of accountability” view. Dislike of “Office Seekers” First, notice that reelection-seeking behavior by politi is generally regarded with disdain and contempt. People widely and often bitterly deplore the fact that politicians “just want to get reelected!” If one has the view that it is precisely the prospect of reelection that induces politicians to do what voters want, then the trait of cravenly desiring reelection should be highly valued, not disdained. We should want to elect those people most desirous of holding office for a long time, because they would be the most inclined to do exactly what we want in order to get reelected and least inclined to risk getting booted by engaging in illicit rent seeking. Note that in Ferejohn’s (1986) “moral hazard” model of the electoral control of incumbents, electoral 60 Control of Politicians control is easier and more efficient the higher the value of holding office for the incumbent (e.g., how much it pays or how much the person values the status), and the lower the reservation value of the incumbent for not being in office. Among other things, this implies that if we fully believed in the “elections as mechanisms of accountability” model, then we should want to pay politicians very large salaries to reduce their temptation to shirk to a minimum (note that the costs would be spread over an enormous number of taxpayers). But such arguments can hardly be found in the public sphere, possibly in part because paying politicians very large sums would encourage exactly the wrong (ypes to run for office - people motivated by financial gain rather than public interest!* Term Limits Second, notice that there is considerable support in the United States for congressional term limits, that the highest office in the United States is subject to a two-term limit, and that a number of Latin American presidents and legislators are allowed to serve for only one term (Carey 1996). These facts would be completely inex- plicable if most people subscribed to the “repeated elections as a mechanism of accountability” view. If you believe that the main point of elections is to offer elected officials a future incentive for representative behavior in the present, then how can you possibly support taking away the principal incentive? On the other hand, if most people understand elections as an attempt to choose a good. type who will of his or her own accord do what they want, then there need be nothing wrong with term limits. In fact, they could even be a positively good thing, if they would help screen out (through self-selection not to run) craven, opportunistic office seekers, leaving instead candidates more genuinely motivated by * Another implication of the pure moral hazard model is that we shoulé want to ‘lect relatively poor people who have a high value for holding onto the office. But {im the United States, atleast, personal wealth is raroly a political lability, and rich candidates are not discounted on the grounds that they will shirk because they won't eare if they lose office. In defense of the moral hazard view. rich politcans might be less likely wo shirk to gain monctary rents since their ‘marginal tility for money is likely to be less than thelr value for office status, (Che's s0 rich he can't be bought”) 61James D. Fearon the desire for public service.’ Finally, note that the argument against term limits that seems to be the most compelling to the USS. public is definitely not that term limits would make politicians unaccountable, but rather that term limits arbitrarily restrict the free choice of who can be one’s representative. They are said to be “undemocratic” for this reason. The Premium on Principles and Consistency ‘Third, notice that voters generally put high value on prin pledness and consistency in candidates and elected officials, even to the point of being willing of forgive some extremism in policy positions if they believe this can be taken as an indication of these qualities. Thus, many people who would have preferred less co! servative policies than Reagan espoused nonetheless appear to have been won over by the sense that he was a man of principle and that they “knew where he stood.”” Further, notice how politicians can face very significant criticism and loss of support if they change ‘a publicly stated position on a contentious issue. This is true even if * Regarding members of legislatures, there is an argument against torm limits fon the grounds that if they are too short, the elected officials will not be able to develop the expertise necessary to allow them to hold their own against permanent staff bureaucrats or lobbying interests. But this argues against hhaving short term limits on representatives, rather than for no term limits at all, Regarding presidents in new democracies, there are some other strong arguments in favor of term limits: First if you can't run for reelection, then ‘YoU can’t use the powers of office to engage in electoral fraud on your own behalf (hough you may stil be able 9 comunit fraud on behalf of your party. Second, if you know that you will be out of office in (say) six years, then you have to worry ‘more about being prosecuted once out of office for illegal acivty you engage in while in office, and the fear of this may then keep you more honest and public-minded while in your one term. Paradoxically, then, term limits for the highest executive office may actualy favor democracy, its continuation, and its responsiveness. ‘There is also the more academic, coordination-problem argument about not ‘wanting tc handicap one’s own representatives with term limits when other states stil do not have them, due to loss of seniority in the legislature (Dick and Lott 1993; McKelvey and fiezman 1992). This has nothing to do with term limits lowering cleetoral accountability, however. ‘To some extent, n the 1996 Republican primary campaign Pat Buchanan tried the same tactic pushed farther. He has espoused and stuck to extreme positions inthe face of a critical reception and electoral fallures, In part bocause this signals that he isa “man of prineiple” who really believes that what he is recommending 62 Control of Politicians they change a previously stated position in the direction of what the public seems to want now! “Walling” is considered a bad thing, even if one waflles toward the median voter. ‘Again, these facts would make no sense if people thought of elections as mechanisms of accountability. If I think of elections as incentive systems pushing politicians to do what I want, then I should be happy and approving when a politician changes position to be more in accord with my views (assuming that I am close to the median, or whatever it is that “the public wants").* And if I ‘want elections to induce politicians to do what is in the public interest, I should be upset with candidates and elected officials who espouse or implement policies not thought by a majority to be in the public interest (i.e., extremist policies or policies away from what the median voter wants or thinks best). I should not approve of them for their principled but unpopular stands. By contrast, if I think of elections as a problem of choosing a competent, like-minded type not easily bought by special interests, then it makes perfect sense to be highly concerned with princi- pledness and consistency. Sticking by a position through thick and thin is a costly signal, since types who are just office seekers and are easily bought will find consistency and principledness more difficult to manage. Last-Period Effects Fourth, if the anticipation of future reelection were the principal factor inducing politicians to vote the way their con- stituents would want rather than for powerful special interests, we would expect to see systematic changes in representatives’ behavior in their last term in office (when they know they will step down due to retirement or a term limit). In other words, we would ‘expect to observe significant last-period effects. In addition, if is best for the country. This line of analysis may suggest an explanation for why |ve tend to see the rise of polley extremists in periods of low trust in government ‘and politicians ~ in such times, the public’s premium on candidates who appear principled and consistent goes up, and this may favor extremists. ‘The point is not that people never approve of elected officials shifting policy positions in thelr preferred direction, but that the common reaction of disdain for inconsistency does not make sense from a purely “elections as mechanisms of accountability” point of view. 63fact saw the reelection incentive as the principal way that elections encourage public policies they desire, then they would be very reluctant to elect a politician for a term known to be his or her last? Neither casual empiricism nor more careful empirical studies reveal strong or obvious evidence for these implications. Regarding casual empiricism, I have never heard anyone claim that U.S. presidential shirking on policy or effort goes up markedly in the second term." And voters are clearly willing to reelect presidents for second terms, and sometimes reelect repro- sentatives who are thought very likely to be retiring or moving on” Regarding more-careful empirical studies, there is a large literature on the question of how much U.S. senators and repre- sentatives vote against their constituents’ preferences, including several studies of voting behavior in the last term (by repre- sentatives who do not die in office).'* With a few exceptions, these "In formal models of elections as a pure moral hazard problem (Barro 1973; Ferejoha 1986), a commonly known last period can cause “unraveling” to ‘occur: anticipating that the incumbent will shirk in her last period, voters ‘would rot reelect, but this gives the incumbent an incentive to shirk in the next to last period, which leads voters not to reclect for this period, and so on - the end result belng that the politclans are completely uncon: strained. Just as in the chain store paradox, this unraveling need not occur if there is lack of common knowledge about which is the representative's last term, even if it is common knowiedge that the representative will eventually dle or rotire. Thus, Bender and Lows (1996: #2) claim that unraveling is a theoretical implication of politicians’ mortality is too strong. One might argue that for presidents the “Judgment of history” isthe relevant sanctioning mechanism in the second term, but of course this is quite different fom the electoral mechanism. During the 1996 campaign it was sometimes said by liberals that in the second term “we will see the real {more liberal] Clinton” ~ that is, that he would shirk, ideologically, against the center. This appears to ‘have boen wishful thinking While one might argue that voters reelect retiring representatives because seniority gives them special powers in Congress, if voters thought reelection incentives were the sole guarantee of good behavior then they would not expect seniority powers to be used on thelr behalf in a representative's last period, ' For a recent review of the lterature on “ideological shirking,” soe Hender and Lou 1996. For studies looking for last-period effects, soe Carey 1996; Lott 1987; Lott and Bronars 1993; Kalt and Zupan 1990; Van Beek 199 190, 64 Control of Politicians studies find no evidence that representatives change their voting behavior systematically in their last term, although they do vote significantly less often (Lott 1987; Lott and Bronars 1993; Van Beek 1991). This bit of evidence is of a different order from most of the previous examples. The previous examples generally concerned the beliefs and attitudes of the public, which might be justified or not. This fact, if true, could suggest that people actually are able to choose “good types” who have similar policy preferences and who are not so easily bought. For this reason it is worth taking a slightly closer look at this literature. If we assume it is correct, there are three possible explanations for the fact that representatives in their (known) last term do not change their voting behavior as compared with their previous records. First, it could be that the electorate is able to sort out good types with similar issue preferences. as suggested here and by Lott (1987). Second, it could be that representatives are. never much constrained by the electorate, so that they vote their ideological preferences all the time, last period or not. Third, in principle it might be that representatives will lose some repu- tational “bond” if they “cheat” in the last period (Barro 1973), such as prospects for higher office, party-guaranteed pensions, or political career prospects for their children. ‘The evidence from the literature on congressional voting seems to favor the first explanation, although more could be done to rule out the second. For the third, there are simply not enough plausible “bonds” available to explain the result for retiring con- gressmen, and those present for some politicians (children’s career prospects in politics, aspirations to higher office) soem to have little effect anyway."* Regarding the second possible explanation, there seems to be no doubt that, when voting, rep- resentatives put some weight on constituent interests as opposed to personal, party leadership, or special interest preferences. The © The major exception is Zupan 1990, which is credibly crtlezed by Lott and Bronars 1993. Lott 1990; Van Beck 1991, See Carey 1996 for a careful analysis that finds ‘that aspirants to statewide office systematically altor their last pesiod con- sessional voting in the direction of the state party delegation, which controls ‘nominations. As Carey notes, however, to few congressmen run for higher office for this to explain the general result concerning the absence of a last-period effect. 65James D, Fearon amount of the weight is hard to estimate."* In an innovative essay, Levitt (1996) uses the voting records of a state's House -gation to estimate constituent interests for senators from that state and finds that senators put three to six times less weight on constituent preferences in voting than they do on “personal ideology” (where this should be understood as personal and probably interest-group preferences). The variation - from three to six times less weight ~ depends on whether one considers only the influence of the state’s median voter, or also those voters in the senator's “support constituency,” which essentially means voters of the same party. While Levitt’s technique seems a big improvement on earlier methods (Kalt and Zupan 1994), the nagging suspicion remains that constituent interests are not adequately estimated, so that the senator-specific effects may contain unmeasured constituent interests. For one thing, in Levitt as in most other studies, the votes used to measure a senator's position are the same for all senators, so that no account is taken of the fact that politicians almost surely put more weight on con- stituent interests on issues that their constituents care a lot about. Brady and Schwartz’s (1995) study of the effect of positions on abortion on election prospects and voting in the Senate tends to support this view."® Levit’s and other evidence in this literature would seem to suggest the following points relevant to this chapter. First, there no doubt that U.S. congressmen “shirk” to some degree in their voting on bills, in the sense that they do not mirror the ideological preferences of their geographical constituencies (whom they are normatively bound to represent)."’ Nor, it seems, do they perfectly ‘See Kalt and Zupan 1994 and the several critiques of their methodology in Public Choice 76 (une 1993), As Bendor and Lot (1996: 78) observe, the hypothesis that politicians shirk completely in every period is inconsistont with the finding that in their last term they vote significant less often (Lott 1987). ‘Another problem with Levit's and Kalt and Zupan’s approaches is thal, average Ideological shirking across senators is effectively constrained tobe zero, ‘that systematic bias introduced by, sa; the greater power of business lobbies cannot be estimated, See Bender and Lott 1996 for this point with respect to the Kalt and Zupan methodology. In Levit, the problem arises from the assumption that the average ideological position of a state's House delegation is an unbiased estimate of voter ideological preferences inthe state © Authors in this literature often bemoan the following problem, What counts as “shirking” depends on which constituency (geographic, electoral, interest sroup?] representatives are representing, and theory does not tell us which is 66 Control of Politicians mirror the ideological preferences of their “electoral support” con- stituencies, although the extent of the average deviation is unclear. Second, diverse evidence suggests that the threat of electoral sanctioning does matter. For example, Levitt (1996: 436-38) finds that senators give twice as much weight to (median) constituent preferences in the year before elections as compared to four years or more before elections. He also finds that senators in marginal seats put considerably more weight on constituent preferences than do senators in safe seats."* Third, it nonetheless appears that voters are to some extent able to use elections to select politicians who have roughly similar issue preferences. Otherwise, it is hard to explain the absence of strong last-period effects coupled with the fact that repr sentatives’ voting does reflect constituent preferences to & non- negligible degree. In addition, some studies provide more-direct empirical evidence of electoral sorting - politicians who deviate more from their constituents’ preferences are more likely to lose reelection bids (Lott and Davis 1992; Lott and Bronars 1993; Wright 1993; Brady and Schwartz 1995; Levitt 1996: 437n23). It appears, then, that both mechanisms of electoral control ~ selection and sanctioning - operate to some degree, although if Levitt (1996) is correct, there remains considerable “slack” in the electoral agency relationship. How Should Elections Be Understood? Are people wrong to think about elections primarily in terms of choosing a good type rather than as mechanisms of accountability? Are they mistaken to disdain reelection-secking politicians, to support term limits, and to put so much emphasis the right one (eg, Bender and Lot 1996; Poole and Romer 1993). From a normative perspective, however, it seems to me that the isu is clear. Repre- sentatives are normatively bound to represent their geographic constiuencies, ‘and not just the people who Voted for them. Thus a representative who votes in ‘accord with the median of her supporters rather than the median of the district is shirking or the median voter in the district has mistaken what Is really in bis ‘own interests. lee is abundant evidence that US, representatives are powerfully motivated by the desire for reelection, and that in consequence they pursue atleast some activities that their constituents desire (e., Mayhew 1974; Cain, Ferejphn, and Fiorina 1987), o7James D. Fearon on principledness, consistency, and the character of candidates and elected officials? These attitudes and beliefs make perfect sense if people also believe that (1) repeated elections do not work well as a mechanism of accountability, because they believe that their ability to observe what politicians do and to interpret whether it is in the public interest is so negligible; and (2) there actually is relevant variation in types of candidates for political office, and these can be distinguished to some extent, either by campaigns or observed performan This is not to say people would be justified in having these beliefs ~ that requires saying whether the beliefs are correct. But neither belief is crazy on the face of it. Consider the second belief ~ that a range of types exists and can be distinguished to some extent. A “good type” for a given voter means a candidate with similar policy preferences, who is relatively honest and principled (hard to buy off, and who is skilled. Certainly candidates vary in their policy preferences and general competence, and almost surely they vary in principledness and integrity. Any variation in types in this sense, and any ability to distinguish them through campaigns or performance in office, would make it reasonable for voters to think about ele least in part as exercises in sorting among types, independent of a relationship of accountability. The more difficult it is to di tinguish types, the more fallible the sorting process - and everyone believes itis highly fallible, since everyone believes there are many corrupt and incompetent politicians. But this need not make it any less rational to try to sort on quality in these dimensions, especially if it is true that elections are not very good as mechanisms of accountability. So now consider the first belief - that repeated elections do not work well as mechanisms of accountability. There can be no doubt that formidable problems are involved in monitoring and evaluating incumbent behavior to make informed judgments about whether to reelect. Voters face a severe agency problem, because their information about politicians’ behavior is inevitably poor and because their sanctioning instrument (reelect/don't reelect) is 50 crude. Voters have neither the time to follow policy debates in Washington nor the training and skill to evaluate conflicting “expert” arguments about what is best. In any event, elected officials can do an enormous amount of business entirely out of public view. Further, when policy is produced by a legislature, it 68 Control of Politicians Is difficult to see how one can hold individual members responsible for it unless one has a detailed empirical and theoretical under- standing of legislative procedure and politics (which even full-time students of Congress do not agree on). Finally, even if the public could better observe and interpret what elected officials do, it is not clear that the electoral sanction is a subtle enough instrument to induce politicians to do what the public wants.” For example, because policy is typically multi mensional, elected officials may be able to play one coalition against another in electoral politics, staying in office while pursuing policies that no majority prefers to available alternatives (Dewatripont and Roland 1992; Ferejohn 1986). Given the difficulty of the agency problem voters face. then. it might be entirely reasonable to imagine that the best available solution is to try to elect good types of candidates, and to view repeated elections as repeated opportunities to sort among types rather than as mechanisms for controlling problems of moral hazard for elected officials. In this view, one votes against an incumbent if economic or other circumstances are bad not in order to give the new officeholder an incentive to work harder or more responsibly, but rather just to try a new random “draw” from the pool of types. In economic jargon, elections may be more about an adverse selection problem - sorting good from bad types that want to mimic them - than about controlling moral hazard. ‘This perspective has been given a formal statement by Banks and Sundaram (1990), who analyze repeated elections as a species of “bandit problem” from statistics.” A voter chooses each period between an incumbent and a challenger, where all candidates are ' Maravall (Chapter 5 in this volume) provides some dramatic examples of how politicians ean exploit the crudeness of the sanctioning mechanism. Why, then, {do we use this simplistic incentive scheme for political offices? Why not te pres~ dential pay to GNP growth (or some weighted average of measures of welfare). as a Saturday Night Live skit had Ross Perot suggesting during the 1992 race? This sounds ridiculous, but It Is Interesting to try to spell out exactly why. Holmstrom and Milgrom (1991) point out that if a job involves many tasks that vary in how easily they can be monitored and evaluated for performance, then ‘making compensation contingent on performance forthe most monitorable tasks will suboptimally direct activity away from the less monitorable tasks. Since politicians’ jobs involve many tasks and responsibilities that vary in their ease of monitoring, tis argument would seem quite relevant See also Rogoff and Sibert 1988; RogofT 1990. For informal discussions, se Lott 1987; Ferejohn 1993, 69James D, Fearon charactorized by an unobserved “compotence level.” ‘The voter observes her welfare each period, which is the incumbent's competence level plus a stochastic error term. The question is when to dump the present incumbent for a return to the pool of challengers. With two candidates, optimal behavior involves a myopic decision rule where the voter dumps the incumbent if per- formance falls below a certain level. This is retrospective voting, but it is aimed solely at finding a good type rather than at giving politicians incentives not to shirk in office. Note that under this conception of elections as pure selection problem, the voter may well want to be able to reelect the same person over and over (ie., not to have term limits), but the reason has nothing to do with electoral accountability. Rather, voters simply want the option of “keeping good ones.” This fits nicely with what seems to be the most popularly convincing argument against term limits, which sees them as restricting democratic choice ~ “if you have a good representative, why force him or her out?” ‘The Interaction of Selection and Sanctioning In the preceding section, I suggested that if voters’ ability to monitor politicians’ behavior in office is very poor, they might reasonably focus on trying to select good types, even if polit vary little in desired qualities. This is a claim about what would constitue rational behavior in a particular situation, and it could be wrong or in need of major qualification. For example, it could be that the inability to monitor politicians also effectively undermines the ability to sort among types. Or it could be that a small amount of variation in politicians’ types would rationally have only a very small effect on voter behavior. One way to investigate and assess claims about what would constitute rational behavior is to consider a simple game-theoretic model of the situation, which I develop next. While neither the model nor its results should be taken too literally, it proves useful for a further purpose as well ‘Thus far, [have described two ways that elections might bring about a correspondence between public policy and what the public desires. Elections might work either as a sanctioning device that induces politicians to choose in the public interest so they ean retain their jobs, or as a selection device that allows the public to choose leaders who will, of their own accord, do what the public 70 Control of Politicians wants." Of course, as the empirical literature cited earlier suggests, there is no reason to think that only one of these mechanisms operates. Even if people tend to think about elections as selection devices, selection can imply sanctioning and vice versa. If people try to select good types at the polls, then this can imply that bad behavior is electorally sanctioned and thus that bad types have an incentive to mimic the behavior of good types. But this mimicry might in turn reduce the ability to select. Clearly, then, selection and sanctioning interact. A simple game-theoretic model is also valuable for analyzing the nature of this interaction. I next consider three variants of a model in which an incumbent politician chooses a policy, and then the electorate, uninformed about the policy but informed about its overall welfare, chooses whether to retain the incumbent. The first two variants depict in stark terms the cases of elections as @ pure selection device and elections as a pure sanctioning mechanism. ‘The third variant combines selection and sanctioning. Technical details are confined to an appendix. T will consider three variants of a simple two-period game between two players, the electorate (or median voter), E, and an incumbent politician, 1. For all three variants, the sequence of events is the same.” 1. The incumbent chooses a policy, represented as a real number z. (dn the pure selection case, the incumbent simply implements a given policy rather than making a strategic choice.) 2. The electorate does not observe the policy chosen, but does observe a measure of its welfare that depends partly on the policy x and partly on random factors. The observed measure of welfare is z = -x" + ¢, where -x" is the electorate’s utility for the policy x and ¢ is a random variable drawn from a symmetric, strictly unimodal probability density function f with mean zero. Thus, the ‘his distinction is implicit in the small formal literature on electoral control under the heading of moral hazard versus adverse selection (see Banks and Sundaram 1993, 1996; Ferejohn 1993). In the public choice literature on “ide~ ‘logical shirking” (eg. Kalt and Zupan 1984), Bender and Lott (1996: 75) label the selection view the “ideology ensuring-performance hypothesis." the model both draws on and tiesto extend the small formal literature on electoral control of politicians. See in particular Austen-Smith and Banks 1989: Banks and Sundaram 1990, 1993, 1996; Forejohn 1986; Harringtor 19938, 19936, Larocea 1997; Reed 1994; and Zielinski 1997. aJames V, Fearon clectorate's ideal policy choice is x = 0, but the electorate cannot tell exactly what the government did because x is not directly observed and the olectorate's welfare level zis a noisy measure of 3. After observing its welfare level z in the first period, the electorate chooses whether to reelect the incumbent or to draw a new officcholder from a “pool” of aspiring politicians. 4. The second period begins, with the new or old incumbent choosing 4 policy y, and the electorate receiving utility ~y? + e”, where &” is another draw from the density f. The game then ends. (From the viewpoint of the first period, all second-period payoffs are discounted by a common factor 5 (0, 1).) A strategy for the electorate will be a rule saying whether to reelect the incumbent as a function of its observed first-period welfare, z. Intuition suggests that a sensible rule for the electorate would take the form of a performance criterion of “cut rule.” For example, “reelect if welfare is at least as great as the level k, and elect a new incumbent otherwise.” In fact, for each of the three variants of the model, the optimal rule for the electorate takes this form, which looks like “retrospective voting” in the sense that Fiorina (1981) and others have argued is typical of how voters decide in practice.** OF course, if voters are rationally secking to influence their future welfare with their votes, then their retro- spective judgment has a prospective purpose. The question then arises of how to choose the performance criterion k to best motivate or select politicians in the future. I proceed to consider is problem for the three variants in turn. Pure Selection In the case of elections as a pure selection problem, candidates vary in their level of competence or skill at producing outcomes the electorate likes.” For simplicity, suppose there are two types = good ones, who are both able and willing to implement the The case where the electorate observes a noisy measure of the policy chosen (= = +2) yelds qualitatively similar result, *1 do not demonstrate this here. See Banks and Sundaram 1996 for a proof of the optimality ofa simple cut rule in a similar model © As in Rogff 1990; Banks and Sundaram 1990. 72 Control of Politicians electorate’s optimal policy x = 0, and bad ones, who can at best implement a suboptimal policy @ > 0. Thus, good types are expected to produce a welfare level of zero on average, while bad types produce ~# on average. Assume further that in the “pool” of aspiring politicians, a fraction a ¢ (0, 1) are good types and 1 - @ are bad types, and that the first-period incumbent was initially drawn from this pool (e.g., in an open-seat race). So in this case of pure selection, the electorate faces the decision-theoretic problem of drawing an inference about the incumbent's competence from its observed welfare level z, and ing if the updated belief that the incumbent is good is her than @, the probability that a new incumbent would be a good type. This is a simple matter of applying Bayes’s rule. In th appendix, I show that when the distribution of “noise” f is symmetric and unimodal, the electorate increases its estimate that the incumbent is competent if it observes welfare z greater than k = #2, which is the outcome halfway between the expected performance of competent and incompetent types. Thus, k =-#/2 is the retrospective performance criterion in this case. If not attained, the incumbent is dumped. ‘Two interesting comparative statics results follow. First, the better the expected performance of the less competent type (the smaller #), the more demanding is the optimal performance criterion (& rises as # falls). For example, suppose that competent types are expected to produce noninflationary growth of (on average) 2%, while incompetent types are expected to produce 0%. Then a rational electorate infers the incumbent is more likely to be a competent type than a new draw would be if it observes growth of at least 1%. However, if an incompetent type is expected to produce growth of 1% on average, then the electorate’s per- formance criterion rises to 1.5%. Intuitively, the more the bad type’s expected performance resembles the good type's, the higher the performance needed to increase the electorate's estimate of competence. This observation will prove important later, in the analysis of the mixed case, Second, making it harder for the electorate to monitor the incumbent's behavior (..e., increasing the variance of the noise ) has no effect on the optimal-performance criterion (k is independent of the density function f). However, worse monitoring does make it harder for the electorate to select good types. The probability that a good type is reelected is the probability that 73James D, Fearon > k, or 1 ~ F(k); the likelihood that a bad type is reelected is t probability that z =~ + e> k, or 1 ~ FU +b). AS the variance of the noise term e increases, both these terms approach 1/2, so that good and bad types are almost equally likely to be reelected. As monitoring improves, by contrast, the probability that a good type is reolected eventually approaches 1 and the probability that a bad type is reelected approaches 0.2 Pure Sanctioning (Moral Hazard) In the case of elections understood purely as mechanisms of accountability, all politicians are the same type. They are all venal, or have policy preferences distinct from the electorate’. Assume, for example, that the incumbent and all aspiring politicians in the model maximize their own utility in a given term in office by choosing x = 1. This might arise either because the politiciens are systematically more ideologically extreme than the ian voter, or because wealthy special interests offer venal politicians various benefits for choosing policies the median voter dislikes, In particular, for the model, assume that an incumbent's utility is given by W~ (1 ~ x) in the first period and W ~ (1 - y)? in the second. W > 0 is a politician's value for oflice independent of policies chosen, which presumably includes pay and perquisites but, for national office, is probably mainly a status value ~ “ego rents,” as Rogoff and Sibert (1988) put it.’” Because the second period is the last, politicians maximize utility in the second period by setting y = 1, which yields them W. For the first term in office, however, an incumbent has to consider the effect of the policy choice on the probability of being reelected, which depends in turn on the performance criterion set by the electorate © Hore and elsewhere, when | equate the quality of monitoring with the variance of e, I mean variance parameterized as in the case of standard bell-shaped ‘curves like the normal and logistic distributions. There are in principle an infinite number of ways to increase the variance of @ distribution, and for some pecullac ways (eg. readjusting probability weight only inthe tails), some of the aims given in the text would be weakened, AS is typical In this literature, 1 will assume Wo be exogenous, even though W ‘an presumably be manipulated through salary choice by the principal. For 4 variety of reasons (some of them worth exploring), representatives’ pay is not much used for incentive purposes, 74 In this case of pure sanctioning, it is important to see that any performance criterion set by the electorate is credible, in the sense that the electorate is willing to reelect or not according to any criterion, Because in this case all politicians are alike (somewhat bad), the electorate is always indifferent between keeping and dumping the incumbent, since it gets the same second-period 1) either way.* And because the electorate is always indifferent between candidates, the electorate can choose the per- formance criterion to motivate the incumbent optimally and have it be credible that this criterion will in fact be employed. As I show later for the mixed case, introducing even a tiny possibility of variation in politicians’ types radically changes thi vary in policy preferences, even a little, then voters are no longer generically indifferent between the incumbent and possible replacements What, then, is the performance criterion that will motivate the incumbent to choose a policy as close as possible to the electorate’s ideal of x = 0? At first glance, one might think that the higher the electorate sets the standard, the greater the incentive for the incumbent to choose a policy in the public's interest. This is mistaken, however. If the electorate holds incumbents to too high a standard for reelection, then, since reelection is relatively unlikely anyway, politicians have an incentive to shirk a lot ~ the marginal impact of shirking on the (low) probability of reelection is small. Similarly, if the electorate sets a low threshold for reelection that is easily obtained, then incumbents have an incentive to shirk substantially. As shown in the appendix, the optimal cut rule &* is chosen to maximize the sensitivity of the likelihood of reelection to shirking by the incumbent. ‘This basic moral hazard model produces a natural comparative statics result concerning the public’s ability to monitor: the worse the electorate’s ability to monitor (the greater the variance of e), the worse the electorate does in terms of policy (the incumbent chooses an x closer to 1). Worse monitoring implies that a given This point ts also stressed by Austen-Smith and Banks 1989, who consider a similar two-period model with moral hazard. It does not seem to be entirely an artifact of having only two periods, however. Voter indifference between candidates obtains as wel In the infinitely repeated case of thls model with no term limits, which I have examined, and in the case presented by Ferelohn (1986), who considers an infinite horizon model where incumbents observe thelr randomly Suctuating marginal cost before they choose “effort” each period. 75James D. Fearon amount of shirking has a smaller effect on the incumbent's probability of reelection, which reduces the force of the electoral sanction, Conversely, as monitoring approaches perfection (the variance of e shrinks towards 0), the incumbent's policy choice approaches the voter's ideal point, x = 0.” The Mixed Case: Selection and Sanctioning As Largued earlier, a number of empirical observations suggest that voters think about elections more in terms of selection than as sanctioning mechanisms to influence future incumbents. The pure moral hazard model just presented (and likewise, that of Ferejohn 11986) allows no place for variation in politicians’ motivations and dispositions, and so cannot capture this dimension of democrati governance that seems to loom large in the minds of voters. While perhaps better, the pure selection model is also crucially deficient. Surely voters do think that some politicians are intrinsically more capable or competent at producing good outcomes than others, and voters are interested in sorting out the relatively competent ones. But everyone recognizes that politicians make important policy choices while in office that are not well monitored by the electorate. Thus Ferejohn and others are absolutely right to say that the voter- politician relationship involves moral hazard in a central way. A theoretical model that captures the voter's decision problem simply but usefully must combine the elements of adverse selection and moral hazard. ‘There are numerous ways that selection and sanctioning might be combined in a game model,2® in part because there are at least © Interestingly, as monitoring worsens, the electorate's optimal-performance criterion becomes less demanding. I have analyzed the infinite horizon version of the pure sanctioning model, and the result appears to obtain there as well, ft least for normally distributed noise. While in the two-period case the probability of reelection is always 1/2, in the infinite horizon case it can either increase or decrease as monitoring worsens (and itis typically well above 12) Banks and Sundaram 1993, 1996 are the only fully game-theoretic election models I know of that incorporate both elements: see the appendix for & discussion of relevant differences. Reod (1994) claims to consider moral hazard fand adverse selvction, but in his model the voters observe the policies nplemented perfectly, so there is no moral hazard in the usual sense of “hidden factions.” Reed's analysis is also incomplete or incorrect. as discussed in the appendix 16 Control of Politi three dimensions along which good and bad types can differ: policy preferences, integrity, and intrinsic ability or competence. 1 will consider a simple formulation that builds on the games just considered, and in which good and bad politicians differ with respect to policy preferences.” ‘Suppose that everything is the same as in the pure sanctioning case described earlier, except that now all politicians initially have ‘an ae (0, 1) chance of being a “good type,” which will mean a politician with the same policy preferences as the electorate (Le. quadratic with the ideal point at x = 0, so that a good types one- period utility function in office is Wx). The electorate now might have an incentive to try to sort out good types rather than just optimally motivate bad types. In fact, as argued informally here ‘and formally in the appendix, even if there is only a tiny chance of finding a good type (a close to zero), it makes sense for the electorate to focus completely on the problem of selection when casting its vote! And, ironically, this is true even though the rationality of trying to select on type at the polls can undermine the efficacy of elections as mechanisms of accountability ‘That it makes sense for the electorate to focus entirely on the problem of selection at election time is easy to see. Since the second-period behavior of all bad types will be the same (they all choose y = 1), at the polls the electorate maximizes its utility by choosing the candidate most likely to be a good type (who would choose y = 0). This holds even if the probability of drawing a good type from the pool of aspiring politicians, a, negligible. "There is an important substantive point here that is probably more general than this simple model. If politicians do not vary in type, as presumed by the pure sanctioning view of elections. then voters are completely indifferent between candidates ~ all politicians will respond the same way to any given set of electoral incentives. But this indifference is fragile. Introduce any variation in politicians’ attributes or propensities relevant to their per- formance in office, and it makes sense for the electorate to focus completely on choosing the best type when it comes time to vote. ‘This result supports and even strengthens the claim made in the previous section, that it would be rational for voters to try to sort » Depending on how wo interpret what gives rise to the politicians’ preferences, types in the model may also be seen as differing in integrity 7James D. Fearon good from bad types of politicians even if there is in fact little variation or it is difficult to discern.®* In the model, the fact that the electorate rationally focuses on type at election time has an additional consequence of interest. Namely, the performance criterion that would be optimal ex ante for motivating bad types of politician is generally not optimal (and thus not credible) ex post. for the purpose of selecting good types. ‘The electorate could minimize shirking by bad types by committing in advance to use the cut rule & that maximizes the ‘marginal impact of shirking on the bad type’s probability of reelection, as in the pure sanctioning case analyzed earlier. But here's the rub: the more effective this performance rule is in motivating a bad type of incumbent to choose a policy close to x= 0, the smaller the practical difference between good and bad types. which makes selection harder. As shown for the pure selection case, the more that bad types choose public-spirited policies, the better the performance the electorate will noed to see to conclude that the incumbent is likely enough to be a good type to be worth reelecting. ‘Thus, the more bad types are expected to try to appear good (ie., the less they shirk the public interest), the higher the per- Jormance standard they will be held to after the fact. And choosing policies more in line with what the public desires will not nec- increase one’s odds of reelection if the public responds by demanding a better performance in order to reelect. ‘This result suggests, then, a possible problem with elections as a mechanism of accountability. The problem is that the electorate cannot commit to keep its standards for reelection constant. If there is variation in politicians’ policy preferences, then the more incumbents serve public rather than special interests, the higher the performance level the electorate will demand for reelection, Its hypothetically possible that in a model with more than two periods, there might be equilibria in which the electorate can commit t keep Incumbents thoughtless good than a new draw would be because failure todo so would lead to “punishment” by future incumbents, This seems intuitively farfeched, in part because such a scheme would probably not be renegotiation-proot ~ if later incumbents were made worse off by punishing the electorate, they would have an incentive to renegotiate to the original terms. Banks and Sundaram's (1993) mode! uses a teigger-strategy punishment by the politicians, but to encourage the elsetorate to drop incumbents if “rewards” fll below a certain level rather than © keep them, 78. Control of Politicians, which may undermine politicians’ incentives not to shirk in the first place. The result can be a sort of “trap” that makes both sides worse off. I show in the appendix that if the proportion of good politicians is not too large (typically, no more than about 1/2), then voters would be strictly better off if they could commit in advance to use the performance criterion that optimally motivates bad types. yp rt can also be shown that because of this commitment problem for the electorate, bad incumbents shirk strictly more in the equilibrium of the mixed case than in the case of pure moral hazard, with the amount of the difference being independent of «, the fraction of good types.” Ironically, then, introducing a small chance that some politicians are public-spirited can make the public worse off relative to the pure sanctioning case, by undermining the ability to electorally control bad types. (Of course, as the fraction of good types grows, eventually the public is made better off by their presence, despite the increased shirking by bad types.) 1 also show in the appendix that, in accord with earlier arguments, as the electorate's ability to monitor becomes very poor, bad incumbents choose their preferred policy (almost) and the electorate’s problem approaches pure selection. At the same time, however, worse monitoring lowers the electorate’s ability to distinguish good types based on performance, so that the efficacy of selection falls as well."© This is not a coincidence ~ bad types are willing to choose very different policies from good types only if the electorate has a hard time telling the difference between them (‘e., if selection based on performance is not very effective). ‘There are thus two senses in which it can be reasonable for voters to focus on the problem of selecting good types. First, since bad types are all the same regarding how they will perform for any given set of electoral incentives, the only basis for preferring © The latter rule is not, however, ex ante optimal for the electorate, because in addition to the sanctioning effect that motivates bad types there Is a “selection effect” ~ the better the performance of bad types inthe frst period, the harder itis to sort good from bad for the second period. Seo the appendix. A perfect Bayesian equilibrium obtains in the model when the incumbent has correct expectations about the performance criterion the clectorate will employ, and chooses a policy with this expectation, In the limit, as the variance of e grows very large, good types’ probability of reelection approaches 1/2 from above and bad types’ from below.James D, Fearon ‘one candidate to another at election time is some variation in type (policy preferences, personal rectitude, competence, etc.). Second, if voters have @ difficult time monitoring whether incumbents chose policies in their interests, then the electoral sanction will not deter shirking by bad types, leaving electorates the option of sorting them out by performance and other measures, albeit poorly. ‘Two other results deserve mention, First, how does the efficacy of electoral selection vary with the voters’ ability to assess whether politicians are choosing good policies? There are two opposing effects. On the one hand, worse monitoring increases the amount of shirking by bad incumbents, which means that the expected performance of good and bad types diverges. This effect increase the voters’ ability to sort good from bad. On the other hand, as monitoring worsens, the probability that a bad type will satisfy the porformance criterion (for any given amount of shirking) goes up, which tends to make selection less effective. In numerical analyses of the case of normally distributed noise, I have found that the relationship between monitoring and the efficacy of election takes the form of an inverted “U.” When monitoring is very good, selection is ineffective (although unimportant) because bad types closely mimic good types. When monitoring is very poor, selection is ineffective because bad types frequently meet the per- formance standard despite their shirking. In between, when voters are moderately able to assess politicians’ choices, the difference between good and bad types’ odds of reelection is greatest. ‘The second result, consistent with those of Banks and Sundaram (1996) and Lott and Reed (1989), concerns a cross section-versus-time-series effect. In equilibrium, all politicians shirk at least as much or more in their second term, due to the last-period effect for bad types. But at the same time, second-term politicians as a group may shirk less on average than first-termers, because selection implies that more good types hold office in the second term than the first.” Thus time series studies of particular representatives would be expected to find increased “ideological shirking” with greater tenure, while cross-sectional comparisons should find less shirking by longer-serving representatives. The empirical evidence is largely consistent with this prediction, % The formal condition for the expected secondperiod policy to be bettor for the electorate than the expected first-period policy is F(-K*) ~ F&*) > (1 ~ V-2k*Ve 80 Control of Politicians although the evidence for increased shirking with greater teaure is slight (for examples, see Kalt and Zupan 1990; Zupan 1990; Levitt 1996). ‘To summarize, this section has posed two “ideal typical” views of how the electoral control of politicians might work, arguing that a good theory needs to incorporate both and also to consider that the two mechanisms interact. Even if voters focus their attention on the problem of selection — and, as suggested by the model, they can have good reasons for doing so even when selection is diflicult = this implies that to some degree elections will act as sanc:ions establishing an accountability relationship. Analysis of the model also produced the unexpected finding that the problem of adverse selection in elections can, in principle, worsen the problem of moral hazard. If politicians vary in their policy preferences and integrity, which they surely do, then (rational) voters will be drawn to choose at election time on the basis of beliefs about type, and ironically this imperative may reduce the flexibility of elections as mechanisms of accountability. In essence, the performance criterion that would minimize politicians’ incentive to shirk is not credible because at the polls voters would want to apply a higher standard in order to sort good types from bad optimally. ‘Though of theoretical interest, I suspect that the practical and empirical relevance of this commitment pathology is small, in part due to concern that the model is unusually spare. Moreover, even in the model as it stands, in numerical analyses | have found it almost impossible to generate large welfare losses for the electorate due to the commitment problem.” Even though shirking by bad types increases when a chance of good types is introduced, the increase is typically quite small, especially when monitoring is poor but even when monitoring is relatively effective. For example, almost regardless of monitoring ability, the electorate does better in the mixed model than in the pure sanctioning case if at least one in five hundred political aspirants is a good type." © tn these analyses I used a normal distribution for ¢ almost surely the claim holds {or any bell-shaped distribution, % Some might say that one in five hundred is too optimistic. Also, the simulation results described all assume that politicians’ value for office Wis at least as ‘great as their value for maximum shirking in a period ~ that is, W> 1. IW < 1, then the commitment problem ean cause nontrivial welfare losses. aJames D. Fearon Conclusion ‘The main justification for freely contested and regular elections is that these are supposed to bring about democracy, or rule according to the will of people. As the editors note in the volume introduction, the tradition of democratic theory contains sur- prisingly little on exactly how elections are supposed to produce this result, or on how well they might be expected to. How, if at all, do elections bring about a correspondence between public policy and what the electorate desires? And under what conditions is the correspondence strong or weak? ‘The main points of this chapter may be summarized as follows. First, following the small formal literature on electoral control, 1 have argued that there are two principal mechanisms by which elections might bring about public policy that voters desire ~ sanctioning and selection, Elected officials might be motivated to choose policies the public desires either because this will help get them reelected (sanctioning), or because the electorate is able to select “good types” who are principled, competent, and share the clectorate’s ends, independent of reelection incentives. This simple point is missed if the voter-politician relationship is understood purely as a relationship of accountability based on the electoral sanction. Second, I suggested that in fact voters think about elections much more as opportunities to try to select good types than as sanctions to deter shirking by future incumbents. A number of empirical observations support this claim. The popular dislike of office seeking, support for term limits, the premium put on politicians being principled and consistent, and the absence of concern about “last-period effects” all suggest that voters think about elections primarily in terms of selection. ‘Third, analysis of a simple model of electoral control in which voters face both adverse selection (the problem of distinguishing ‘good from bad) and moral hazard (imperfect monitoring of agents Wwho might shirk) suggested two reasons why it may be reasonable for voters to focus on selecting good types. Once at the polls, the electorate can’t commit to ignore type in favor of implementing an ‘ex ante optimal sanctioning scheme ~ variation in type is what is relevant to voter payoffs at the moment of choice. Also, if the electorate’s ability to monitor what politicians do is poor, then the force of the electoral sanction is weak and voters are left with Control of Politicians tte] ala Ce eae of getting the public policies they jesire. Finally, the mechanisms of selection and sanetioning inevitably interact, with consequences for the efficacy of democratic rule Even if voters focus on selecting good types. this implies accountability, since bad performance in office will be sanctioned when it suggests the incumbent is a bad type. In the model examined here, where voters could only distinguish types of politicians by observing their performance in office, the electorate’s ability to monitor determines the interaction between selection and sanctioning. Good monitoring induces bad types to act like good ones, so selection becomes difficult but also less important. Bad monitoring leads bad types to shirk more, which makes selection possible, but also makes it a noisy and fallible enterprise for the voters.” ‘The importance of monitoring for making elections somewhat effective instruments of popular rule in the model accords with common sense (though I hope the strategic logic described provides stronger and more developed foundations for the ‘common sense). Almost surely, politicians are most inclined to choose policies and other actions that the public desires when the probability of exposure for failure to do so is highest. The standard liberal observations about the importance of effective media and an informed, interested public follow immediately. As Downs (1957) stressed, however, voter's incentives to be interested and well informed about politics are far from strong, a point amply confirmed by empirical studies of what most voters know about politics. The preceding arguments about monitoring, selection, and sanctioning are all based on a model in which people assess politicians’ types by looking at noisy measures of their performance in office. As noted earlier, there are other, less attention-intensive ways of trying to assess types. For example, ® As throughout, “monitoring” refers here to the electorate's ability te assoss ‘whethor politicians aro choosing policies in the public intrest. I have not dis tinguished among the several reasons why voters may have trouble making this assessment ~ stochastic relationships between policies chosen and ostcomes ‘observed (what is modeled here), lack of knowledge about what policy isin one's fown best intorest (Harrington 1993a), asymmetries of information between politicians and voters concerning policy (Larocca 1997), and so on. Differ- fences among these may have significant consequences for accountablity and responsiveness. 83D. Fearon how does the politician look and sound when giving a speech? What sort of person does he or she seem like? It is an interesting and open question as to how well voters can select “good types” based on this sort of information, and quite relevant to the question of how well elections serve democracy. Appendix. Pure Selection Derivation of the electorate’s decision rule proceeds as follows. Let az) represent the electorate’s updated belief that the incumbent is a good type, conditional on observing welfare level z. Bayes's rule states (in odds-ratio form) that the posterior odds that I is a good type, a’(zV(1 - a’(2)), equals the prior odds, a/(1 ~ a), times the likelihood ratio, flzyfi# + z). (The likelihood ratio derives here from Pr (z ~ 0 + eV/Priz ~ ~4° + e). Thus, a(z) > aif and only if flz) > flé + 2). Intuitively, the posterior belief that the incumbent is a good type increases if it is more likely that the electorate would observe z if the incumbent were good than bad. ‘The optimal performance criterion is then defined by the welfare level k that satisfies f(k) = flé? + b). As indicated by Figure 2.1, for f such that f(z) = 0 only at z= 0, there is a unique cut rule k that satisfies this equality. Further, when f is symmetric as well, we can specify the performance criterion exactly. It must be k = ~2/2, since ft) = ft-k) = fl + 8) ‘To see the claim in the text about monitoring, consider a logistic distribution F(z) = e**/(1 + e**), where a > 0 parameterizes variance (greater a implies higher variance). Observe that since the optimal cut rule & <0, lim, 1 ~ Fk) = lim, 1 - FCA) = 12, limo 1 — F\k) = 1, and lim. 1 - F(-#) = 0, as claimed. The same holds for a normal distribution. Pure Sanctioning ‘The electorate’s optimal cut rule &* is found by considering how the incumbent would choose given any particular rule k, and then choosing k so as to induce the smallest choice of x possible. As argued in the text, because the electorate is indifferent between all candidates in the second period, itis optimal for it to implement any cut rule. Ra Control of Politicians 4 7 + 4 F(a) 3 o Ls a f@) ; 2 nu Ls ° T % ete Figure 2.1. Pure selection Given a k, the incumbent chooses x to maximize u(x) =W ~~) +0(1- Fat +k), a where 1 ~ Fix? +) = 1 ~ Pre-a? + 6 < k) = 1 - Priz < &) is the probability that the incumbent will be reelected given x and k. Notice that the right-hand side of (1) implicitly incorporates the fact that the incumbent will choose y = 0 in the second period, by subgame perfection. ‘The first-order condition is w(x) = 2(1 ~ x) ~ OW flr? + 2x = ©, which yields eepelczt Sat += oe @ Provisionally assuming that the second-order condition is also satisfied, equation (2) implicitly defines the incumbent's best response x as a function of k; call it x(k). The right-hand side of (2) is decreasing in x, so that E gets the smallest x possible by choosing k so as to maximize the left-hand side, given x(k). The assumption that ft) is strictly unimodal and has mean zero implies that fl) reaches its maximum value at f(0). Thus, to induce the 85James D. Fearon minimum possible value of x, E chooses k* such that x(k")! + k* = 0. Letting 2* = x(k*), we can solve directly for x* from ex 0) ae a ‘Thus, x* = (1 + 6Wf\0)) and k* = ~x*8 As the variance of a normal or logistic /(-) increases, /(0) falls, increasing the equilibrium amount of shirking (x* increases), and making the performance criterion less strict (k* falls). (It is worth pointing out that the solutions for x* and k* hold for any strictly unimodal distribution for ©.) It remains to check that the second-order condition is satisfied at x* and &*. This condition requires that =-2-2OW1 f(a? +b)+ 222/22 +b) <0 co) 0, f(0) > 0, and FO) wis) at x* and k*, which is true since x(k")! + & =0. The Mixed Case: Selection and Sanctioning The game is solved for pure-strategy perfect Bayesian equilibria (PBE) in which the electorate uses a cut rule in deciding whether to reelect the incumbent. Specifying strategies is sim- plified if we immediately apply subgame perfection to the new or old incumbent's choice of a policy y in the second period of the game: For all prior histories of play, good types maximize their utility by setting y = 0, and bad types by choosing y = 1. With this understood, a pure strategy for the incumbent in the first period ‘amounts to a choice of x € R for each type. A pure strategy for the electorate is a function ss(z) © (0, 11, which gives the electocate’s decision to keep (1) or drop (0) the incumbent for each possible observation z ¢ R. Attention is restricted to functions of the form se(z) = 1 if and only if z > k” Let s = (ty, x, &) represent a strategy triple where the first two elements are the good and bad type’s first-period policy choices and the third is the electorate’s cut rule. A system of beliefs for the electorate is a “© See Ranks and Sundaram 1996 for a proof of the optimality of such cut rules in ‘a similar model 86 Control of Politicians function c(2) ¢ 10, 1], which gives the electorate’s posterior belief that the incumbent is a good type after having observed welfare level z. A pure-strategy PBE is then defined as an s and a system of beliefs «’(z) such that each player's strategy is optimal given beliefs at every one of the player's information sets, and the electorate’s beliefs are formed using Bayes's rule whenever possible. Working backward by considering the electorate’s choice of whether to keep or drop the incumbent, observe that E's expected utility for reelecting the incumbent is a’(z)0 + (1 ~ @(2)-1 -(1 ~ @'(z)), while the expected utility of drawing a new incumbent from the pool is just -(1 - a). Thus optimal choice for the electorate entails keeping the incumbent if a'(z) > a. In odds ratio form, Bayes's rule says that az) | flape2) a Tae)” figez)i-a so that E keeps the incumbent if flxz + z) 2 flrs + 2) and not otherwise. Further, B is indifferent when it observes z ~ &, implying that in equilibrium & satisfie Sx +k) = fix + bs 6) Next, consider what choices of x, and 2, would be optimal for the incumbent in the first period, given the expectation that the electorate is using the cut rule k. The bad type’s maximization problem is exactly the same as in the pure sanctioning case considered earlier, so condition (2) again implicitly defines a bad type’s best response x(k). Good types choose x to maximize W— x? + 6(1 - Fx? + &)W, which has its first-order condition -2x - ft? + k)W2x = 0 satisfied uniquely by x = 0. (That this is a maximum is apparent from inspection of the objective function.) So good types have a dominant strategy of choosing their ideal point x regardless of 8, W, and k. ‘Thus, condition (5) reduces to f(k) = fis + k), which implies, since fl) is symmetric and quasi-coneave, that in any equilibrium ksOand kana. © This last equality says how the electorate’s performance criterion k responds given the policy choice expected from bad 87James D. Fearon Figure 2.2a. Poor monitoring (high-variance noise) incumbents.*! As in the pure selection model, the better the bad types" expected performance (smaller x,), the higher the level of welfare E needs to observe to conclude that the incumbent is more likely a good type than a new draw would be. But of course the cut rule & also determines which 2, is optimal for the incumbent through condition (2). A PBE requires that the bad type’s choice of x, imply a cut rule k (through condition (6)) that in turn induces bad types to want to choose x, (through n (2)). Formally, 25 = V-2k from (6), and substituting for x in @) yields the condition that implicitly defines “interior” candidates for equilibrium &*'s: 1 V-2E 1 Sb o Figures 2.2a-c graphically depict the determination of such k's in three cases. The following claims are apparent from inspection of the figures: (a) There exists either one or three interior solutions to (7) (omitting nongeneric cases involving a tangency). (b) If ft) © Nowe also that ks 0 and x + £20 imply that in any equilibrium good types will Dereelected with a probability of atleast 1/2, while had types are reelected with probability at most 1/2 88 Figure 2.2¢. Good monitoring (low-variance noise) is normal or logistic, then for large enough variance there is a ‘unique interior solution that approaches k = ~1/2 (thus, x = 1) as variance increases. If the interior solution is indeed a maximum, 89James D. Fearon then this implies that as monitoring worsens, bad types’ choice of policy ultimately approaches x = 1, or total shirking, (c) For interior solutions, shirking need not decrease monotonically as monitoring improves (.e., variance of f decreases), although for small enough variance there will exist an interior solution that involves almost no shirking (Figure 2.2c). Text show that neither 2: can be part of a PBE, +50 that if'a PBE exists it involves at least one of the interior solutions to (7). At.x,=0, w'(0) = 2 — 26W flE)(0) > 0 for all ke [-~, =}, so bad incumbents always have an incentive to shirk at least a little. For x= 1, observe that w'(1) =-25W f(1 + &) < 0 for all finite k, which means that the incumbent wants to choose x) < 1 for any finite cut rule, And since neither k =~ (sure reelection) nor & = (certain replacement) can support an equilibrium ~ in both cases E would optimally deviate to k = -1/2, ex post, since 25 = 1 would be expected “any pure-strategy PBE must involve an interior solution to the bad type’s first-order condition given earlier. For cases like Figure 2.2a, where there is only one interior solution, this must be a maximum since w(0) > 0 and w'(1) <0 for all finite k. For cases with three interior solutions to the first-order condition, I next show that the “middle” solution corresponds to a ‘minimum, and thus is not a candidate, Using the facts that at a solution to (7), x* +k kf) = ft-R) = (1 V-2kVG-2k AW), and f*CR) =f"), algebra applied to the second-order condition (4) can be shown to imply that w(K) < 0 if and only if Pye ® 2a oW Differentiating the right-hand side of (7) with respect to k gives 1A(-2k)75W) as the slope of this function. Since at a “middle” interior solution to (7), f(k) > 1AC2k)"76H"), it follows that the second-order condition (4) cannot be satisfied. So we can say that the “middle” interior solution to (7) cannot support a PBE, and that at least one of the other two solutions (in this case) must support a PBE. Without specifying functional forms, we cannot say which one or both supports an equilibrium. ‘© jn simulations using a normal distribution fore, | have not found a single case in which multipte equilibrium obtain. ! have been unable to show that this is necessarily the case ‘ontrol of Politicians ‘This completes the derivation of equilibrium in the mixed case. {conclude with a comparison of the mixed and pure moral hazard models that demonstrates and elaborates several of the claims about the models made in the text. Comparison of Mixed and Pure Moral Hazard Models In the pure moral hazard case, the best the electorate can do is to induce politicians to choose the policy tq that satisfies conditions (3). This policy is defined implicitly by ar Zan OW £00) ‘Suppose now that we introduce into the pool of candidates a very small fraction (a close to zero) of good types of politician who have policy preferences similar to the electorate or who are uncor- uptible by interest group bribery, as in the mixed case. Writing (7) in terms of x = v-2k, where 2 is the equilibrium choice of bad types in the mixed case, we have Seed SO)= since fiz) is maximized at f(0) and x, > 0. It follows that regardless of how small « is, bad types of politicians shirk more when there are some good types than when they are all the same (ty > qu): This Is a “pathological” result ~ introducing a sufficiently small chance that there are public-spirited politicians who will actually lower the public’s utility, essentially by making the public unable to commit not to select on type at the polls. The imperative of selection then reduces the flexibility of the elections as a sanctioning instrument and leads to suboptimal motivating of bad types of politicians. In more technical terms, lower hemicontinuity of the equilibrium correspondence fails at a= 0; there exist equilibria in the a= 0 case that are not the limit of any sequence of equilibria as a approaches zero. This result is in marked contrast to Banks oti ct the (1 play oe indeterminate jin the mixed case. a 1James D. Fearon and Sundaram (1996), who consider a closely related model of elections that combines moral hazard and adverse selection. They find that the equilibria of their game with a small amount of type heterogeneity (adverse selection) “converge to equilibria of the ‘pure moral hazard’ case,” so that “adding a ‘little bit’ of uncertainty about agent types will only move the agents a little bit away from their myopic actions” (1966: 25). A related implication is that the electorate “can . .. never lose from the introduction of ‘better’ types of agents” (1966: 5). Neither result holds in the model considered here, and it is natural to wonder why. ‘The crucial difference appears to be that my model ends in two periods, whereas Banks and Sundaram (1996) consider the case of an electorate with an infinite horizon and politicians who are constrained to serve at most two terms." In pure moral hazard models with an infinite horizon and any commonly known, finite term limit, the electorate cannot commit to condition reelection on performance, and, in consequence, incumbents are completely uncontrolled in any subgame perfect equilibrium (“unraveling” occurs). By contrast, in a two-period, pure moral hazard model, voters can credibly promise to reelect incumbents who perform well. This allows equilibria in which politicians do not shirk completely in the first period, as shown here and by Austen-Smith and Banks 1989. I have considered the infinite-horizon, two- period term-limit case of the mixed model analyzed earlier, and find that Banks and Sundaram’s (1996) result is reproduced there introducing a small chance of a good type leads to a small improvement over the total shirking equilibrium of the pure moral hazard case. The interesting question is whether the commitment problem discovered here would appear in an infinite horizon model with no term limits and both adverse selection and moral hazard. Banks and Sundaram 1993 satisfies these conditions, but there the politicians’ choice variable is not policy but effort (as in Ferejohn 1986), and adverse selection is introduced by varying politicians’ marginal costs for effort. It is easy to see that even in the two: period model considered here, if politicians vary only in their * Another difference that may be significant is that ideal-point preferences are not covered in the large class considered by Banks and Sundaram, which is formulated more with “effort” in mind rather than policy chotees. Ideal-point preferences fal Banks and Sundaram’s Assumption 6 92 Control of Politicians marginal costs for moving policy toward the median voter, then the electorate will be indifferent among all types in the second period and thus optimal sanctioning will be ex post credible. 1 suspect this is why no commitment problem appears in Banks and Sundaram 1993. (If correct, this would also indicate that the choice of “effort” or policy as the politician's choice variable is nontrivial.) Thus far, I have been unable to analyze the infinite-horizon, no-term-limit case of the model with politicians who vary in policy preferences. For the reasons given in footnotes 29 and 32, | would conjecture that the commitment problem will appear here as well as in the two-period case. A final significant point about the mixed model is that moral hazard is not necessary to generate the commitment problem pathology discussed earlier. Consider the same model but with perfect monitoring ~ assume the electorate observes the policy choice x directly, and so can condition the incumbent's reelection on it. Consider a rule for the electorate that says, “we will reelect if you choose our optimal policy, x = 0.” A bad type of incumbent will want to mimic a good type and choose x = 0 provided that W = 1+ 8W2 W+ 0, or 6W> 1. And in this event, the electorate is willing to reclect if it sees x = 0, since its posterior that the incumbent is good is then a’(x = 0) = a, implying indifference. So in this case adding a small chance of good type has no effect - the electorate gets an expected payolT of approximately 0 + 5(-1) whether a= 0 or is very close to 0. Suppose, however, that 6W < 1, In this case the best the electorate can do when all politicians are bad types (a = 0) is to employ the rule “we will reelect if you choose a policy x <4,” where is defined by W- (1 -2)' + W= W, or é= 1 - VW. 2s the policy choice that leaves a bad type indifferent between choosing £ to get reelected and shirking completely and not getting reelected. But if some politicians are good types (> 0), the electorate will infer that the incumbent is a bad type if it sees é > 0, which implies that it should dump the incumbent in favor of a new draw who has some chance of being a good type in the second period. Thus what was an equilibrium when a= 0 collapses, leaving a radically different and unique PBE in which bad types shirk completely, choosing x = 1 and being tossed out of office for sure. In this case, then, introducing a tiny chance of public-spirited politicians com- pletely undermines elections as a mechanism of accountability. 93James D. Fearon it is not difficult to show what these arguments go through if we consider an infinite horizon model with no term limits and no ‘moral hazard.) This last example also suffices to show that Reed's (1994) analysis is either mistaken or incomplete. Reed’s model is ‘essentially identical to that just analyzed, except that he considers case where incumbent's ideal points are uniformly distributed. Reed solves for the # that is optimal for the electorate ex ante, but never asks if ex post it would be rational for the electorate to implement this rule in all states of the world. The example just given suggests that it almost surely is not, and that if it is under uniform distribution, this would be an exceptional case. One claim from the text remains to be demonstrated: that for small enough «, the electorate would be better off if it could commit in advance to use the performance criterion k* that optimally motivates bad types (defined following equation (3). 1 see this, we write the ex ante expected payolf for the electorate in the mixed case as follows: up = (0+ TF(RY-(1~a))+ (1 axy—g-+ Fxg + ee —1)+ (L- Fag + XD), where the & and x, are equilibrium values in the mixed case. Algebra simplifies this to ups = (1 ark xf + [1 ~ a F(R) ~ FUR). Note that the term in brackets captures the payoff improvement due to the possibility of selecting a good type for the second period. In a similar fashion, the electorate’s ex ante expected utility for commiting to the rule &* defined for the pure moral hazard case is found to be (1a xiy +f ~a( 5 A(R*))). where tun <2 follows that u," 24> dfa(F(-K)~ FUR) -(5-(K")))} the bad type’s policy given k*. From algebra, it > ug" when For a close enough to 0, this necessarily holds, since xf ~ x2, >0 and & and k* are bounded between 0 and -1/2. In simulations with a normal distribution and 6W = 2, the statement typically holds for all a less than about 1/2. For a> 1/2, the utility loss from 94 Control of Politicians suboptimal selection in the second period is greater than the gain from better motivation of bad types in the first period. Austen-Smith, David, and Jeffrey Banks. 1989. “Electoral Accountability and Incumbency.” In Peter Ordeshook, ed., Models of Strategic Choice in Politics, 121-50. Ann Arbor: University of Michigan Press. Banks, Jeffrey, and Rangarajan Sundaram. 1990. “Incumbents, Challengers, and Bandits: Bayesian Learning in a Dynamic Choice Model.” University of Rochester. Mimeographed. 1993, “Adverse Selection and Moral Hazard in a Repeated Elections Model.” In William A. Barnett, Melvin J. Hinich, and Norman J. Schofield, eds., Political Economy: Institutions, Competition, and Rep- resentation, 295-313. Cambridge: Cambridge University Press. 1996. “Flectoral Accountability and Selection Fffects.” University of Rochester. Mimeographed. Barro, Robert. 1973. “The Control of Politicians: An Economie Model Public Choice 14: 19-42. and John Lott. 1996. “Logistator Voting and Shirking: A. ow of the Literature.” Public Choice 87 (1): 67-100. Brady, David, and Edward Schwartz. 1995. “Ideology and Interests in Politics of Abortion in the U.S. Senate.” noice 84: 25-48, Cain, Bruce, John Ferejohn, and Morris Fiorina. 1987. The Personal Vote. Cambridge, Mass.: Harvard University Press. Campbell, Angus, P. Converse, W. Miller, and D. Stokes. 1960. The American Voter. New York: John Wiley and Sons. Carey, John, 1996, Term Limits and Legislative Representation, Cambridge: Cambridge University Press. Dowatripont, M., and G. Roland. 1992, “Eeonomic Reform and Dynamic Political Constraints.” Review of Economic Studies 59: 703-30. Dick, Andrew R., and John R. Lott. 1993. “Reconciling Voters’ Behavior with Legislative Term Limits.” Journal of Public Economics 50: 1 14. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row. Ferejohn, John. 1986. “Incumbent Performance and Electoral Control.” Public Choice 50: 5-25. 1993, “The Spatial Model of Elections." In Bernard Grofman, ed., Information, Participation, and Choice, 107-24. Ann Arbor: University of Michigan Press. Fiorina, Morris P. 1981. Retrospective Voting in American National Flections. New Haven: Yale University Press. 95James D. Fearon Frank, Robert. 1988. Passions within Reason: The Strategic Role of the Emotions. New York: Norton. Harrington, Joseph. 1993a. “The Impact of Reelection Pressures on the Fulfillment of Campaign Promises.” Games and Economic Behavior 5: 71-97. 1993b. “Economic Policy, Economic Performance, and Elections.” American Economic Review 83 (1): 27-42. Holmstrom, Bengt, and Paul Milgrom. 1991. “Mulli-Task Principal-Agent Analyses.” Journal of Lau, Economics, and Organization 7 (1): 24-52. Kalt, Joseph, and Mark Zupan. 1994. “Capture and Ideology in the Economic Theory of Politics.” American Economic Review T4 (June) 279-300. 1990. “The Apparent Ideological Behavior of Legislators: Testing for Principal-Agent Slack in Political Institutions.” Journal of Law, Economics, and Organization 33 (April: 103-31. Key, V.0. 1966. The Responsible Electorate. New York: Vintage Books. Kishlansky, Mark. 1986. Parliamentary Selection: Social and Political Choice in Early Modern England. Cambridge: Cambridge University Pross. Larocea, Roger. 1997. “Informative Representation.” University of Chicago. Mimeographod. Levitt, Steven D. 1996. “How Do Senators Voto? Disentangling the Role of Voier Preferences, Party Affiliation, and Senator Ideology.” American Economic Review 86 (June}: 425-41. Lott, John. 1987. *Political Cheating.” Public Choice 52 (March): 231-46, 1990. “Attendance Rates, Political Shirking, and the Effect of Post- Elective Office Employment.” Economic Inquiry 28 (1): 133-50. Lott, John, and Stephen Bronars. 1993. “A Critical Review and an Excension of the Political Shirking Literature.” Public Choice 74 Wecember): 461-84, Lott, Jchn, and Michael Davis. 1992. “A Critical Review and an Extension of the Political Shirking Literature.” Public Choice 76 (May): 125-49. Lott, John, and W. Robert Reed. 1989. “Shirking and Sorting in a Model of Finite-Lived Politicians.” Public Choice 61 (1): 75-96. Manin, Bernard. 1997. Principles of Representative Government. imbridge: Cambridge University Press. Mayhew, David. 1974. The Electoral Connection. New Haven: Yale University Press, McKelvey, Richard, and Raymond Riezman. 1992. “Seniority in Leg- islatures.” American Political Science Review 83: 1181-1206. Poole, Keith, and Thomas Romer. 1993. “Ideology, ‘Shirking,’ and Repre- sentation.” Public Choice 77 (2): 185-96. Heed, W. Robert. 1994. “A Retrospective Voting Model with Heterogeneous, Politicians.” Economics and Politics 6 (March): 39-58, 96 Control of Politicians Rogoff, Kenneth. 1990. “Equilibrium Political Budget Cycles." American Economic Review 80: 21-37 Rogoff, Kenneth, and Anne Sibert. 1988. “Elections and Macroeconomic Policy Cyeles.” Review of Economic Studies 55: 1-16. Van Book, James. 1991. Public Finance Quarterly 19 (4): 444-56. Weber, Max. 1978. Economy and Society Vol. 2. Berkeley: University of California Press. Wright, M. B. 1993. “Shirking, Political Support, and Agency Costs in Political Markets.” Public Choice 76 (June): 103-24. ‘Zielinski, Jakub. 1997. “Parties and Elections in Emerging Democracies.” Paper presented at the 93rd annual meeting of the American Political Science Association, ‘Zupan, Mark. 1990. “The Last Period Problem in Politics: Do Con- sgressional Representatives Not Subject to a Reelection Constraint Alter Their Voting Behavior?” Public Choice 65 (May): 167-80. 97